- Hey, it's me, Destin. Welcome back to the second channel. I have thought for years that when I'm ordering a
computer or something like that, if I just buy the most expensive computer then obviously it's
going to be the fastest and it's gonna be the best for me, right? In one particular case right now, editing 4K footage, this is a 4K video, it takes forever for me
to render the footage and I lose sleep and I'm not with my family
as much as I want to be. It's a problem. It's an operational
deficiency I have in my life. So in an effort to fix this, I decided to go to a custom
computer company in Seattle named Puget Systems. They're the only people that
have a website dedicated to application-specific hardware. Meaning if I'm going to
edit in Adobe Premiere, I go click the Adobe Premiere button, they show me the exact computer I need. If I'm doing 3D CAD, they
show me the computer I need. Twitch streaming, a
gaming box using Unity, whatever it is, they
have a computer dedicated in terms of configuration
for that specific process. I found this fascinating. They have hardware specific solutions for software specific needs. It's pretty cool so I asked
them if I came up there and did a little tour behind the scenes, maybe talk about Moore's
law, stuff like that, if we could do a trade where
I'll make a video for that in exchange for some hardware. So they hooked me up
and I really enjoyed it. You know, you don't have to
buy things from Puget Systems but I found the way they do business to be really interesting. So I hope you enjoy this
behind-the-scenes look at Puget Systems. These guys are not marketers, they are just computer
builders and computer testers. It's really interesting. Let's go check it out. So I flew to Seattle. There's a guy named John
about to pick me up. He owns a company called Puget Systems and they make specialty computers
for certain applications. We're gonna go meet with them. What's up, man? How's it going? Can I throw stuff in the back? - [John] Yes. - [Destin] Awesome. You doing alright? So you own the company? - [John] Yeah. - Okay, so this is John. He owns a company called Puget Systems and you're basically the answer to everyone's computer problems, right? - Well, not all the problems. (laughs) - [Destin] So what is the
purpose of the company? You make specialty computers for specific applications, right? - Yeah, we have to understand how our computers are being used to understand that software
just as much as the hardware so we can make the right
computers for the job. - Oh yeah? So editing is a nightmare for me. You gonna hook me up? - Yeah, yeah. I mean, so that's pretty common for us. With people working on
high-speed footage or 4K, 6K, it takes a lot and so we have to understand
exactly what you're doing in order to build the
right hardware for it. - So what's your story? Like how did you start
a company like this? This sounds like a
company that would start because you get frustrated at something. - Yeah, well, you know
I started it in college and frankly it started off as like let's try to cash flow college. But as I got into it, I saw that there was so much
need that could be filled. So many people wanted to
just sell you hardware and not actually solve the problem. And so where we found our
path has been with, yeah, solving the problem. What are we actually trying to do? The computer is a tool to get a job done. - Right. Yeah, that makes sense
because back in the day, it was, what, Dell and Gateway
and Apple, that was it. When did you start the company? - I started in 2000. - Okay, so yeah, that
was about that timeframe. Where that's how you bought computers. So now it's changed. They still, you know, you open a magazine and there's one option. You know, like you can build a
computer on a lot of websites but you do a totally different thing from what I can understand. - Yeah, because we don't really ask what hardware do you want? We ask what are you trying to do and then we can tell you
what hardware you want. - Oh my gosh. When you run a test from a rocket, you don't ask if you want a
high-speed camera or a radar, you ask what is the data
you're trying to accomplish? - Right, what is your goal? - [Destin] What is your goal? And so you start from there and then build the hardware
based on what they say. - Right, right. - Why is that so revolutionary? - I don't know, it's shouldn't be. No. - That makes perfect sense. Okay, cool. Where'd you go to school? You said you went to
University of Washington? - Yeah, I went to the U-Dub. I was studying physics there and I started this company, I think it was in my third year of college and this company kind of took
over my life at that point. - Oh, so you didn't finish. - [John] No, I didn't finish. - You pulled a Bill Gates. - I did everything except the
labs, that's what I have left. But I don't think I could
just go back at this point and finish my labs. - Oh man. Surely they could let you. Is it like a computer lab, what was the? - It was physics labs. And so we were studying, oh, I don't know. We were doing computer design actually was one of my favorite labs. We were bread boarding
components together, making computers from scratch. We were doing tests of
verifying the speed of light and things like that with
lasers going through mirrors. So it was a lot of fun but the company took up a
lot of my time at that point, building that, and so yeah. Went to do the computer thing. - Alright. I'm excited to see your real lab now that you got kicked out
of lab or quit school lab, I'm ready to see your real lab. You're kind of tucked away here. It's kind of a nice little area. - Yeah, we like it. We spent a long time finding this place. - [Destin] That's awesome. We believe that computers
should be a pleasure to purchase and own. They should get your work
done and not be a hindrance. I think I'm at the right place, dude. (laughs) Holy cow. - That's where I thought
you'd be the most interested. So we're set up for us to be in there for a lot of the day. - [Destin] Oh, so he's testing stuff? - Yeah, yeah. Hardware or software. - [Destin] Everything? - This is from the whole approach of it has to get that job done. That room is where those
questions are answered. - Okay, I want to go to that room. - [John] Okay. (keys clacking) - Okay, here we go, tour time. So we're gonna start with just
the overview of what you do? - Yeah we'll just at a high level go through the whole process. - [Destin] Okay, cool. - Overall our mission is to know what the right hardware is
for each application, right, and so that has to start with testing. And so in labs is where
all of that begins. - [Destin] Is this William? - It is William. - [Destin] Okay, yeah. I interacted with William. Sounds good. - [John] Matt on the right. - [Destin] What's up? Hey, I'm Destin, yeah. What's up, William? How's it going? - Good. - [Destin] Yeah, it was a pleasure working with you via email. So this is what like when we were talking, you were in a lab like
this the whole time? - Yeah, I was just
sitting right over here. - [Destin] Really? - Yeah. - [Destin] So what are you doing? These are... - [William] We are trying to
play with Nvidia's NVLink. - It's stuff that no one
knows if it works or not and no one has an answer, so
trying to find out the answer. It's really hard to find an answer. - [Destin] Is that a new card? - These are not particularly new cards but the NVLink's been around
for a couple generations on the Quadro cards as a
way to link cards together and let them communicate
and even share memories. So if you got two 16 gig cards, they'd have 32 gigs together. If it's set up right in
supported applications and now on the latest RTX cards that you saw a few of down there, they switched over to using NV link on the mainstream G-force cards but-- - You're saying a bunch of words and I'm sure they're important. I'm sure all of these words are important. (laughs) So break it down
for an idiot from Alabama. - So-- - It's a way of linking
together multiple video cards and turning them into one
more powerful video card. - [Destin] You're basically
taking these video cards and putting them in parallel or in series? - [Matt] Parallel. - [Destin] Parallel? - [William] More like in parallel. - But it's parallel in a way that the software doesn't have
to be specifically coded for, which is a big deal because a lot of software
is not coded very well to take advantage of multiple items. - Well, and what's weird about NVLink, the original implementation did have to be application specific and it had all sorts of weird caveats and we just went through a
bunch of steps to set it up. But they supposedly are implementing it on the mainstream new cards and everyone's excited about this being some awesome new thing but we think from everything we've
been able to see so far, they're not actually implementing
the full version of it. They're just using it
as a sort of simple way to do not the full deal. They're not going to share memory, it's not gonna have
all these cool features that the Quadro version has, it's just gonna be used for gaming. - [Destin] And so they're
probably gonna go forward and charge extra for another version. - Well, they want you to
pay for the Quadro cards is what they want. Nvidia, for professional uses like this where you need a ton of video
memory or multiple cards to work together well, they want you to pay the thousands
of dollars for the Quadro instead of the hundreds
of dollars for a GeForce. - Am I understanding this correctly? So you guys get the hardware
and then you test it. Like you're full-on
basically test engineers. You figure out how it all works and then whatever the manufacturer says they're giving people, you can tell if it's real or not. Is that pretty much what your business is? - Yeah, we can tell what
it's gonna be good for. So in here we have to, so we obviously have all the hardware. We have all this new hardware. We also then have to pair
that up with the software and then we can see, okay, is this technology for
linking video cards, is that gonna help video editors? Is it gonna help scientists,
is it gonna help engineers? And we can then look and for all these
different software packages that people run, we then will know what hardware will optimize their work. - Really. So you just buy the stock
stuff off the website or do you have deals with these companies that they'll give you a card? - Oh, it depends. Yeah, a lot of times you'll have to go off the shelf and buy. A lot of times with
with Nvidia, Intel, AMD, Gigabyte, Corsair, we'll have
really good relationships where they'll support
the work that we're doing by providing us the hardware. - [Destin] So this is like the
old Consumer Reports model, only modern day and for nerds. - Yeah, yeah, I suppose so. - [Destin] That's pretty
much what it is, right? - Yeah, yeah, and it's all about, it's not about like is this
reliable or not or that. It's all about the job
that has to be done. So it's applied to we're
running a Photoshop benchmark or we're running a
DaVinci Resolve benchmark or William might be running
SolidWorks or Autotask. - [Destin] So explain a benchmark to me. So I assume it means there's this one task that we're gonna perform
over and over and over with different pieces of hardware and see which one works the best? - Yeah, for that very specific task. So you have to engineer your
benchmark very carefully because it's easy to fall into a trap of saying this one
benchmark is going to be the one-size-fits-all
measurement of everyone's work and that's just not true. You might apply different
filters to your videos than some other creator
would apply to their videos and so we have to benchmark
all these different facets of the software and build up this profile. So if you go on our website and look at the things we publish, it's just torrents of charts and graphs and you have to know
which of the ones apply to your specific work. We have to speak your language because it's not about the frequency and gigahertz and cores, it's about a blend effect
in a Premiere timeline, how does that benchmark
on certain hardware? - [Destin] Yeah, you've
got a YouTube channel and I've gone to it and my problem is setting up
my Adobe Premiere workflow and so it says you need
to have your application on this hard drive and you need to have a
solid-state drive over here where you're rendering. So do you optimize the workflow and then apply different graphics cards? - It's a two way street. I mean, we have to know
what hardware is capable of providing this best speed but then we also have
to work with end-users to educate on how to get the most out of the hardware they have. - So I assume, this is what you do, that you're becoming
increasingly more popular with the YouTuber community, people that are doing
really hardcore editing but they're just like at their house. Is that true? - Yeah, well, and as
everything goes from 1080P to 4K, 6K, 8k, everyone's
kind of in the same boat where they're hurting for a compute power, hurting for performance. And it's not just YouTubers. Right now there is a lot
of independent film studios that back in the day you had to be a Pixar to do something cool, right? You had to have these
huge rendering farms, it had to be this huge organization. And with how much progress
we've seen in hardware, it's opened up to the masses. So now you have these small
independent film studios, five people, 10 people,
that are doing amazing work but they don't have that
huge infrastructure. They need an advocate
to tell them do this, use this hardware, use it in this way. Here's how to set up your timelines. - [Destin] So do you feel like you're kind of like training
wheels for people like me? You can help me. You know, I don't know
anything about graphics cards, right, other than a 1080
TI is better than a 980. - We think that it's a
disservice to folks like yourself to have to, you shouldn't
have to figure out technology, how computers work, in
order to do your job. And so we view ourselves as a bridge between your world and
the technology world where we have to understand
the hardware and the software so that we can have a talk
with you in your language, have a talk with Intel in their language, and then we can put together something that meets both needs. - So in the interest
of not making this feel like a commercial for your company, who else is doing this in the space? Because I asked you if I could come talk about all this stuff and you agreed to build a computer for me, so let's just be clear about that. There's an exchange happening here. But who else is doing
this or is this just you? - Well, I mean, I can't say that there's nobody else doing this but the reason that we
focus so hard on this is that we see there's such a void, that not a lot of people are doing that. You'll see companies that will take on like big data center workloads and you'll see a lot of places
that'll do gaming machines and consumer machines, but this world of workstations is like where server meets consumer and it's kind of this blend in the middle. And we don't see a lot
of people going after it and that's why we are. - [Destin] So who are the
hardest people to serve, the video creators, the
engineers, or the scientists? And the answer is not engineers. (laughs) - I think that actually
the scientist end is both the most challenging
and the most rewarding because these are people that are basically writing their own code or on the forefront of
research of use cases that we haven't even realized yet. And so with that, it's
not just taking a piece of software off the shelf
and optimizing for it, it's actually working with the
scientists with their code, their writing, and see how how
their application works out. So it's very hard. There's no one-size-fits-all, it's working with each
person one at a time, but it's really cool work. - [Destin] Really? - Yeah. - So how do you, I mean, back
in the day you would go to, like if we're talking about earlier, you'd go to Dell or Gateway or Mac and just pick out your computer and do it but you're tailoring
hardware for the application. That's what you're doing. - And well, I'm sure we'll
talk about this a little later with what's happened with Moore's law. One-size-fits-all computing,
those days are over. And so now it's tailored
hardware for the application. So just going to Dell and
buying something off the shelf, you might get the right
thing but you don't know. You have to have an expert
to really kind of tell you what the right thing is. - If I tell you a story about something, will you not think less of me? - Okay, sure. - So I bought a Bitcoin miner because I wanted to understand it, okay, and then ASICs came along. - Yeah, exactly, perfect example. - Application Specific
Integrated Circuits. So I bought an ASIC. I know I will never make my money back in this particular scenario but I wanted to understand the
difference in the hardware. So what you're saying is that
as we get more complicated with our computations
for scientific research or modeling for SolidWorks
or whatever it is, you need application specific
hardware just like the ASIC. - And the ASIC is like
the full realization of the work that we're kind of doing because with an ASIC
for like Bitcoin mining, it was a very defined problem and so you could design hardware
that that is all it does is mine those coins and it's exceptionally
fast at it because of that. Conversely, when you have
a general-purpose computer off the shelf, it can do
it, just not very fast because it wasn't purpose-built for that. Now where we live is
anywhere in between that where for video editing or CAD design or scientific simulations,
it's a diverse enough problem that we can't make an ASIC for it because that's way too
expensive of hardware for something that's too
diverse of a problem. And so we have to take
this off the shelf hardware but kind of make it more ASIC-like in that we're designing
it for a specific purpose. - [Destin] So a normal
off-the-shelf type solution, you know, the processor
will bounce back and forth between the CPU and the graphics card and in the workload and
where stuff's happening. Is that true? - Yeah, well, you have to know-- - I'm really stretching myself here so does that make sense? - I guess in the most basic way, it's all about bottlenecks. Knowing where your bottleneck is and then attacking more kind
of your budget on buying a PC towards wherever that bottleneck is, yes. But that's all conditioned
by how your software works and how you use your software. - [Destin] Wow. So this changes monthly, right? So like William, you just got, these cards just came in, right? - Yeah the 20 80s, the
GeForce RTX series cards are brand new and actually
speaking of purpose-built stuff, they're the first time
Nvidia to my knowledge has started putting purpose-built hardware in their video cards. They've got some cores that are dedicated just to ray-tracing, which
their aim is video games there, being able to do reflections and lighting and shadows more accurately. But we're hoping that
in the upcoming releases of ray-tracing software
like Octane and V-ray and those sorts of things, that hardware could also be used there. We'll have to wait and see if that happens but that's the expectation. That's one of the things we'll be testing. So we've done testing now on the versions that don't support them. New versions come out, we test again, and we see, oh, there's a
huge increase of performance. They must be using those cores
now or that sort of thing. We can write about it. And one of the interesting things that John was sort of mentioning, I'm sure other companies
are doing this testing but we're one of the few that I'm aware of that actually publish this information. Because it'd be real
easy if you're Dell or HP to do all this sort of testing
and then keep it secret so that you can in theory
design good computers. - Manipulate the market as well. Like you say oh, I'm gonna. I've seen that in the ASIC
market for example, right? Like the ASIC guys are making
these circuits for mining and then they're holding
them themselves, using them, and then they're only
releasing certain things at certain times in limited quantities. I know you guys aren't
into manipulating markets or anything like that but you want to decrease the probability that other people can do that. - [Man] Well, that and we just want to share the information. - We're all about being
free about it actually. So all the testing that's going
on right now in this room, the results will be
published on the internet. Anybody can come and read it. - [Destin] How do you
make money doing that? Like I'm looking at William. William breathes in words
like ray-tracing and Octane and so how do you pay William and then make this profitable? - There's certainly a
contingency of people out there that love to read all of
the articles we put out and become those technology experts. But there's also a lot of
people that don't want to have, you shouldn't have to
become a technology expert reading dozens and dozens of articles in order to figure out what thing to go off the shelf and buy. So the service we provide
is if you get it from us, all of that is baked in. Or if you don't want to buy
from us, that's cool too. You can go to our website, you could read all the articles, and you could get to the
same conclusions we come to but you would have to become
as much an expert as we are. - [Destin] So basically you have black magic computer ninja
warriors for hire, right? And people buy from you and they have access to their brains. - Exactly, yeah. And there's nothing secret or
proprietary about what we do. It's just we do the work and
we're passionate to share it. - And it lets us also
prove to people, you know, you don't have to come to
us and take our word for it that we're selling you the right thing. We have the data to back it up, you know? Whereas again, if you go to Dell or HP, maybe they've done testing,
maybe they haven't, I don't know, but unless they can
actually show you that data, you just have to take their word for whatever the best hardware is. They might be right, they might be wrong. We can prove it. - [Destin] So are you a
computer test engineer? - I guess. - [Destin] What's your title? - Solutions research and development is. - We got to work on that. We got to work on that, man. We got to work on these titles, man. I was thinking black magic
computer ninja warrior was pretty good so... - Black Magic is the
company that makes some of the software we test, actually. - [Destin] Oh yeah, I guess
they are, aren't they? - Fusion and stuff. - Okay, so not black magic. Like dark arts or something like that. - [Matt] We'll get sued. (laughs) - [Destin] Yeah, fantastic. I noticed you have some
RGB stuff going on here. Is that is that important to you? - It's really hard to not get on high quality components these days. A lot of the really nice, like motherboards especially,
are actually made for gamers and they put all this
RGB and stuff on there, really high quality stuff, but yeah, they make it all flashy and most of it can be turned off at least. - [Destin] What are your
thoughts about that? So do you think like when people
are doing genome sequencing or whatever it is they're doing, you don't think they need the disco ball going on in the back? - I think it depends on the person. I think some of those scientists
are like all about that. I mean, they grew up in the 70s and stuff. But I mean, our big thing is like we do like they're just honestly
kind of boring boxes because they just sit on
your floor or on your desk and they just do what they need to do. And so yeah, we don't
do a whole bunch of RGB or even like liquid cooling
stuff was big a while back. So no, we don't do that stuff as much. If it's on there, great, but oftentimes we just turn it off. - [Destin] So I'm seeing
Lightroom go crazy in the background. Is this a Lightroom benchmark? - Yeah, yeah, so that's
benchmarking Lightroom right now. DaVinci Resolve, that's
the color grading stuff. It's not the nicest looking actual project because we're just testing things like on this one power
windows and color grading and effects and things like that. But yeah we do a whole suite on-- - [Destin] So this is automatically going through the benchmark? - Yeah, so he wrote the
software that's automated a lot of this testing. Because oftentimes for the
kind of work we're doing, there is no industry
benchmark so we make one. - Yeah, because... And some of these we can release publicly. Like I think our Photoshop
benchmark is public, you can just go and download it. Some of these ones are a
lot more temperamental. So like Premiere Pro is probably one of the most temperamental ones because-- - [Destin] Amen. - It's not. (laughs) The software itself actually doesn't crash for us very often. It's that it's not made to be automated in the ways that we're automating it. So like even like (mumbles) our timeline. - [Destin] How are you
doing this right now? Like you recorded mouse
movements and clicks? - Essentially, yeah. It's hard-coded mostly but it's different on every application. Like Lightroom has pretty decent scripting so we can mostly use built-in scripting. This is DaVinci Resolve. It does not but we know like
where things are supposed to be and like we have to duplicate
some steps to make sure that it's always consistent. And then this one actually, we can't read what the FPS is. It'll start playing in a bit. We can't read the FPS but
we can take a screenshot of just that area and
then log all of those and then do pixel checksums to figure out exactly what that value was. So it's annoying very often. - [Destin] You do a pixel
checksum and use that. That's like a low-budget
way to use computer vision. That's exactly what he's doing, isn't it? - The precursor to machine learning. It's just really, really
poor machine learning. - [Destin] Oh wow. - But yeah. But we wouldn't be able to
do all the testing we do without this automation. I mean, some of our
benchmarks like the Lightroom and Photoshop stuff
takes about 20 minutes. DaVinci Resolve takes about 40. But like some of them, Premiere Pro, it's about a six hour benchmark is I think what we're at right now. - [Destin] So what are you,
what's your metric then? Is your metric time? - It depends, it depends. On a lot of like Lightroom
stuff like exporting photos, yes, time. DaVinci Resolve, Premiere
Pro oftentimes is FPS or something. - [Destin] What do you mean by FPS? - Frames per second. So if you have-- - [Destin] Yeah, I understand that but like how can I use that to tell me how well it's performing? - Essentially it'll let you know. So like in Premiere Pro,
I know you run into this, you'll start playing your timeline and it gets super choppy. If it was a 24 FPS media and we are able to play it at 24 FPS, then you know there's no choppiness. But if it's only getting 12 FPS, then you're basically
skipping every other frame. So that's basically just our metric for is your system able to
keep up with different things. And again, like John was saying before, it totally depends on what you're doing. So like for Premiere, we test I think it's like
a dozen different codecs. We test with a bunch of like
heavy effects or transitions. Lumetri Color and all that kind of stuff. - This is why I'm here. So I run the Phantom. It's a Phantom high-speed camera and we crunched the data
down and we have a video and then we usually apply some noise filtering software to it. - Yeah, actually. - [Destin] You have it? - Yeah. - [Destin] The file I sent you? - Yes. - Yes. You gave it to him? Thank you very much. This is why I'm in this room. So yeah, I have a problem where, I mean, it takes forever to edit this stuff, dude. And so-- - Yeah, I looked at it. It's cool stuff. - [Destin] Which file did I
send you, the cannon shot? - You sent a close-up cannon,
it was like 1280 by 720 and then farther out 1920. - [Destin] Yep. And so you guys like Adobe Creative Cloud because they change every
month or whatever it is now. - There's updates coming soon
that are gonna be even worse because like they won't support Windows 7 and Windows 8 anymore on the next version and all this other stuff,
so things change constantly. - [Destin] So do you have
to rerun the benchmark? - Oh yeah, yep. That's how we have a job still. - [Destin] Really? - We have a job because we have to constantly redo this
stuff over and over and over. - So it's any time the hardware updates or any time the software updates. And when you put those
two things together, it's a near-constant clip
of something's changing. - [Destin] So do you give
Adobe themselves feedback and tell them what they're screwing up? - Well, we don't put
it that way, but yeah. - [Destin] So let me put it this way. Do you tell them when they screw up? - We do have feedback back into it. As with on the hardware
side, on the software side, we have varying levels of relationships with different vendors. Some are very, very responsive and love to speak with us. Some we're still working on them, still working on that relationship. Adobe specifically is very,
like every team is different so I don't know what teams
you're in, most met with, but like Lightroom, After Effects, those are different teams we deal with than dealing with Photoshop. - [Destin] In general, do people recognize that you're trying to help their software
run more effectively? - I think in general. - [Destin] And so people
are pretty cool about it? - Oh yeah, oh yeah. - [Destin] So here's the deal. So that's it just playing, right? - Mm-hm, live playback. - [Destin] Oh man. Yeah, that's live playback. - Yeah, that has Lumetri Color on it because normally it's very flat. - [Destin] Right. Yeah, so the kinds of things that happen, you see this graininess here? That's noise that happens because the Phantom doesn't
have an optical low-pass filter and so we'll go in there and run this software called Neat Video. - Yeah, actually we downloaded the trial. - [Destin] Did you really? - On there because we
don't have a full license, but yeah, that's that thing. And yeah, man, that thing just, I don't think I even set it up correctly 'cause it's not really doing very much but yeah, it chunks. But we looked at it and it's because it's
only using a single core. So like if we have... - [Destin] You mean the software itself? - The software itself will
only use a single core on the CPU. It jumps around so it's
kind of hard for you to kind of see there. That one core is super high and so that means that you have to tailor, if this was the only thing you
cared about was Neat Video, you would want to get a CPU that like core count doesn't even matter. You just want that one core
to be as fast as possible. Now it's really hard when
you get into like Premiere because, well, you're also
applying Lumetri Color or you're doing transitions or you're playing back other
footage at the same time, so it's the balance between
all of those different aspects. But there is also just some stuff that you just cannot make
run on multiple cores. It just is not possible because like if you're
doing a whole bunch of math, you have to rely on the first bit of math, the result on that to run the second bit and then the third bit. So some stuff is just like that. CAD modeling is a good
example of one of those. And this might be that way. I mean, without being a
developer for Neat Video, there's no way for us to know. - [Destin] So it's possible
that getting a racehorse machine might not speed up that
one particular process? - No, probably not. And in fact this CP that we have in here, this is the highest end
consumer CPU from Intel, it's 18 cores. This is probably at not as
good as a, say, a $500 CPU. Like today the 8,700 case,
it's just only six cores instead of 18. - [Destin] For this
particular application. - For this one thing. - [Destin] Really? - Now it's not gonna be like massive. It'll be like, total guess
here, 10 or 15% faster, so like this live playback, it's still gonna be chunking along. But you know, that might come into play if you're doing like I know
you often like render previews to get that to go. That might be able to finish 10% faster and maybe that would be worth it for you. - So let me understand what just happened. You sell fancy machines
with a lot of cores in them. - Sometimes. If it's right. - And I came to you asking for a machine and instead of telling
me the expensive machine is what I need, you just told me the cheap machine's what I need? - Depends on exactly what you are doing and the balance between everything. Again, because you're doing a
lot of different stuff, yeah. And again, like it's all about like okay, you can sacrifice 5%
here but you get 15% here or you sacrifice 10% here
and you get 30% here. It all just totally depends
on what you're doing. - [Destin] This is my one
specific application here. So is this how you work
with all your customers? They say this is what I'm doing, like I sent you that file
and that video software and said this is my problem. This is what you do? You just sit in this room and you-- - Doing customer level stuff, we do that for some of our customers depending on what they're doing
and all that kind of stuff. But I mean we try to make our testing be as specific enough that
we don't have to do that for every customer. Because if we already
know the answer, great. Like if you come to us and you just say I'm
doing in Premiere Pro, I work with red 8K footage, and I just basically do
Lumetri and basic trims. We already have the answer. We can tell them exactly
what they need to do. - [Destin] Plop, plop, fizz,
fizz, this is what you do. This is the recipe. - Yep, exactly. - And we have to constantly keep learning. Like now that we've
tested in your Neat Video, we now contain the
knowledge in our company that Neat Video uses one or two cores and so next person that comes
to us to ask about Neat Video, we have the answer now. - [Destin] So how do you document all of this stuff you're learning? - Mm-hm. (laughs) - [Destin] So is it the problem that every engineering company has? - Probably. We try to get it out to
at least the consulting and support staff because they're
the customer facing folks. You know, so that sales guys can help when a customer asks and
the support folks can help if a customer maybe from a year or two ago is suddenly using Neat
Video and asks about it, they need to know as well. But it can be a challenge to make sure everybody
gets the right information and that we keep track of it so that we hire somebody
who wasn't in the meeting where we told everyone, we need to make sure
they know that as well. So it does get a little tricky. - We actually just
yesterday rolled out a LMS, a learning management system,
that we use internally in the company and so we
are publishing training, internally facing training, about video editing, engineering,
and scientific computing that all of our employees go through. - [Destin] Really? - From the person that sells the computer to the person that boxes up the computer. We have that knowledge
throughout the company. - [Destin] So you said
that happened yesterday. - We rolled it out yesterday. We haven't finished. - [Destin] Oh, okay. So what problem do you
think it'll solve for you? - First and foremost, it's
one of our passions is the customer experience we provide and one of the ways that
we do that is we need to know the language of our customers. And so we believe that even beyond just like a technical standpoint, understanding the
workflow of our customers and having that empathy
and having that vocabulary will elevate just the whole
experience we can provide. - [Destin] So you're creating a culture. - Yeah, a culture that's not about tech. A culture that's about the
work that the tech is doing. - Got it, yeah. 'Cause that's how I found you guys. We were trying to crunch this software and it was bogging down
and we were sad a lot and so we started looking for solutions. And on your website you have
a specific build tailored for Adobe Premiere and
that's how we found you. So I mean that's your
bread and butter, isn't it? - Yeah, yeah, absolutely. That's the way we approach it with all of the applications we target. - [Destin] That's awesome. - I think it'd be cool
at some point to talk, like you wanted to talk Moore's law, and I think we should do that because it kind of tees up a lot of what we're talking about
here because back before-- - [Destin] Do you have a cleanroom? - [Man] Not really. - [Destin] Is this the lab? We're in the lab? - Yeah, I mean, this is a lab. But I could take you out the warehouse and you can see like how we
actually put things together and do quality control and
image computers and whatnot. - [Destin] Yeah. Let's talk Moore's law over there. - Okay. - [Destin] Does that work? - Yeah, absolutely. - [Destin] Yeah. So you just got CPUs laying around? - Yeah, well, right now
because I'm swapping through. That one I'm doing a bunch
of Lightroom testing on for a thread ripper as
the big platform from AMD but we got to do these
benchmarks over and over and over to test all these different CPUs. So it'll be one, two, three,
four, five, six, seven CPUs from Intel, I think
like four CPUs from AMD. But yeah, you got to run
that benchmark, swap the CPU. - [Destin] So there a difference, do you take a statistically
significant number or you just test one CPU per benchmark? How do you do it? - The CPU usually doesn't change too much so if you buy the 79 80 XE from Intel, generally every single one is
gonna be within a few percent. Not enough to really-- - [Destin] And you know that
because you've looked at data? - Yeah, yeah. Well, and a lot of these
benchmarks actually we're starting to roll them out to our
entire production process. So in production right now, I know they are already running
our Photoshop benchmark, so the same thing we run up here, it's run on every single
computer out there. And that we have some like rendering ones. Eventually we'll have
After Effects and Premiere and Lightroom so that actually
each customer machine, if they tell us that
they're running Premiere, we do this big six hour
benchmark in Premiere to make sure that their
computer is running exactly in line with what we
expect for what they're doing. - [Destin] You mean before you ship it. - Before we ship it. - [Destin] Really? - Because, I mean, if
there's off on weird things. Like a Windows update
could do something funky. A video card driver update
could do something funky. And if we catch it here, that means that our customers don't have to deal with it themselves. - [Destin] Oh, wow, because
it has to do with the, I don't know what the word is, the symbiosis between the
software and the hardware? - Yeah and if something changes-- - [Destin] For the specific application. - Yeah, and if something changes. And we have a great
example of those RTX cards that we were talking about earlier. Very, very new and often
times there's little bugs when something is very,
very new and it was true. In Photoshop, if you enlarged an image or used smart sharpen,
you got image artifacts. Like built into the image,
not like on your display only. And so we actually know
that so that we would know to tell customers like
either hold off on this or this will probably be
fixed by a driver update and we'll let you know when it is or do you want us to hold your machine until there is a fix,
and we can figure it out. - [Destin] Either a
software or hardware fix. - Yeah, yeah, whatever it takes to fix it. But we have that knowledge. We can tell the customer that so they don't get it, immediately
start using Photoshop, and there's this weird bug thing going on. They've already been warned. - [Destin] Can I ask you a question that's gonna make John feel weird? - Yeah. - So clearly your brain
is the most powerful thing in the company and your brain
is the most powerful thing in the company. Why do you still work here? - I love it here. I've been here for 17 years. - [Destin] Really? - Yeah, yeah. 'Cause John's my older
brother so I started-- - Oh, there you go, there's the answer. (laughs) There's the answer. - No, yeah, I started working with John, I mean I was still in high school. I think I was a junior in high school and I worked through
Puget all through college and after college. Now it's been a while since college. - [Destin] Did you graduate? - No. - [Destin] Oh okay, got a family
thing going on here, yeah. - Brothers, yeah, we got
a family thing going on. No, no, he got closer than I did but it just got to the point where Puget, it's just so fun here. Like I mean especially when you get to how long we've been here. Like I've been through production, I was production manager
for a while, or repair, QC. And so you worked somewhere long enough, especially in a small company, you get to really tailor what
you do to what you enjoy. So right now, I like this. I like writing benchmarks,
I write testing stuff, I like writing. - [Destin] That's cool. - [Matt] It's awesome. - [Destin] What about you, William? Why are you still here? - Well, not related. I've been here a long time
too, 13 years as of yesterday. - [Destin] Is that common
with people that work here? - Relatively. I mean, we certainly, I think we see probably more turnover in production and support but a lot of the people
have been here a long time. The head of our consulting department was the first actual employee
after John himself. We've got quite a few folks who've been here a long time. It's a good environment, a good set of folks to work with. - I didn't mean to put you on the spot. I was just wondering. - No, no problem. - Because I mean how do you keep knowledgeable people like that? I mean, that's a management
thing, how do you do that? - Yeah, well, I mean I think
we aren't a cold large company. You know, we're a family and we are one that are unified in mission that we all want to do the same thing which is we all geek out over the work that our customers are doing and how much technology
can do for the world and so we just feel like we're
playing that kind of role in the world. - [Destin] I've seen a lot of people that have companies like this and they get so good at something that people from Alabama find them when they're looking for a
machine and the solution, right, like me. And so when thousands of people like me start coming to you, how do you scale that and keep that good-old-boy culture? - That's a good question. We specifically say that we do
things here that don't scale and so we have to be very specific about what problems
we're willing to tackle and we keep that focused
very, very narrow. And if we start to get too big, then we kind of have to
prune the tree a little bit and we have to go and
decide what are the things that are most important to us that we're gonna solve this problem? - [Destin] Okay, so you're not just going after every opportunity
that comes down though. - No, we're very specific. We have a number in our mind
of how many different software and hardware platforms
we think we can support and that's how many we go after. - [Destin] Really? But you could hire more people and bring more software
and hardware platforms into the fold. - We could. - [Destin] Do you do that or? - Well, no, no, because we feel like the relationship is what makes everything different here. The relationship amongst
people here in the company and the relationship
we have with customers. What we do talking with
a specific customer and tailoring something to their needs, that doesn't scale and so we don't have a
burning desire to scale. - Interesting. I would ask Matt if that was true but he's your brother, he's
not going to tell me the truth. - It's true. (laughs) I mean, there's even good examples like we used to sell laptops. And you know, eventually we just decided that we wanted to focus on workstations so we dropped laptops. It was not an insignificant
amount of our business. - [John] It was 40% of our business. - 40% of our business. - [Destin] You just dropped
40% of your business? - Yeah. And we're happy. - [Destin] Really? - We're happy, laptops were not fun. Laptops break a lot more than desktops. It's harder to get high quality laptops that we can really add value to. I mean, because we could
just resell Dell laptops but we wouldn't actually be
adding any value in there. So unlike workstations,
all this kind of stuff that we can really fully
tweak absolutely everything, laptops we couldn't and so
we just decided to drop them. Yeah. - [Destin] Wow. Was that a hard decision? - Oh well, I mean yeah, when you look at the numbers on the page, it makes it for a hard decision. But then you look at the experiences and it's not a hard decision. The experience wasn't as good and if we're tailoring
hardware to use case and we can't tailor the hardware, then what are we doing? We're doing one-size-fits-all
computers too. So that's that's why we focus on desktops. - [Destin] Got it. So you don't want to be
a production type house. You want to be a specialty shop. - We like that we're big enough that we can actually get the job done, that we have a statistically
large enough sample size and we're big enough that
partners are interested in working with us but we want
to find that happy medium. We don't want to be, yeah, turn into a big commodity volume seller. - [Destin] Interesting. So you operate on margins
or how do you do that? - Yeah, yeah. I mean the way that we
keep all of this funded is when we sell the computers. Yeah, they will be more expensive than you'll find from
Dell, not obnoxiously so, but the way that we position it is you have the right computer for you so it actually ends up
being faster for the dollar than if you were to go to somewhere else, buy cheaper hardware but it not be the right hardware. - [Destin] Got it. Well, if you can keep me
from rendering at 3:00 a.m., then that'll work for me. Yeah, cool. Sweet. How did the Titan stack up? - It depends on what you're doing. Again, software specific. In a lot of things, this
2080 TI is performing about the same as the
Titan V at half the cost. So DaVinci Resolve is
actually a good example that's up for color grading and stuff and yeah, these two cards, same exact performance, half the cost. So yeah, we're going to sell
that one, not the Titan V. - Yeah, for Resolve and
there's other applications like the Titan V is faster in rendering and in scientific stuff. Especially there, Don really likes it, one of the guys in labs has got it. He's the one with the PhD around here, the rest of us didn't finish college. Makes up for it. But the Titans support 64-bit precision, they call double precision, and so he loves it for a lot
of the scientific workloads where the other cards can't even, either can't do those calculations or do them horrendously slow. - Well, you know, I'm not gonna lie, I'm so intimidated by you. Really every time you start talking, I'm like he's gonna say words. He's gonna say words that are real big. I don't know what they mean. This is your world, isn't it, man? This is your, this like, how many benchmarks are
you running right now? - I've only got one thing actually running and we had some stuff going over here but we're trying to get these
NVLink stuff to cooperate. - We probably do, when we
publish like a whole series for like these new video cards, we probably do maybe 10
different software packages for what we're testing,
somewhere around there. - Yeah, it depends on what comes out. You know, if it's a new
CPU or a new video card and where it's targeted will determine what stuff we're gonna run and what stuff we're gonna publish. Nobody's chomping at the bit to hear how the next mainstream
CPU does in GPU rendering because that's the video
card's what matters so we don't run GPU
rendering tests on that. But yeah, it really
depends on the hardware we're looking at and
again software updates. When a new package comes out, we try to run all the hardware that we think might be appropriate there to see what is the best. And Adobe stuff especially
has been changing in how it behaves over the
last several years, right? - [Man] Yeah, yeah. They've added more and
more GPU accelerations. The video card is becoming
more and more important but only to a certain point. - [Destin] What is that? - [William] That's one of the new NVLinks for the 2080 cards. It hooks onto-- - [Destin] Does it link
them together or something? - [Man] Yes. So yeah, that's what we
were talking about before. - [William] Little cover here. You put it on two cards. - [Destin] Oh, and you gang them together? - Yeah.
- Yeah. And the question is how
does that actually work because Nvidea says kind of one thing and then these other
companies are making claims about it being able to do other things, so that's what we're trying
to figure out right now is this is super new stuff and figuring out whether it
actually does those things that the companies were claiming. - [Destin] Is it just copper? I mean, is this just a jumper? - [Matt] Yeah, pretty much, yeah. - [William] There's a little bit of other stuff in there, I think. - [Destin] So did you take it apart? - No. We don't tend to do a whole lot of that. There's a lot of websites
that do tear downs and stuff. We care more about whether
it works and what it does than how, if that makes sense. Because, well yeah, what
we care about in the end, we want to make sure we
convey to our customers is what it does for them. They don't probably care what
it looks like on the inside but they care if as some
companies are claiming, it allows you to pool the memory on two video cards together. That could be huge for
a lot of applications but Nvidia doesn't say it does that and so we're trying to
kind of test and see whether these other companies
are correct in their claims or whether they're using information that they found somewhere else and it's just kind of
misinformation getting passed along. We like to solve that. You know, break-- - [Destin] Misconceptions, yeah. - Love doing that. - [Destin] That's awesome. - The Internet's not always right. - [Destin] What? Well, Smarter Every Day
videos are always right. - Oh, those are always right. (laughs) You gotta find the gems. - I'm joking, I've messed up. There's two things I can think of. I made a mistake on a dragonfly video and there's another mistake, I don't remember what it was but yeah. - [Matt] Yeah, I mean,
we make mistakes too. - You lose sleep over them. - I mean, all you can do is
when you make those mistakes, you put out a correction. We even have on a lot of our articles, there's a set like date
where they become obsolete and it even puts a big
banner at the top saying that like this information
is from two years ago, it might not be correct anymore. So take it with of grain of salt and here's our newer articles if you want to see the newer stuff, 'cause yeah, it changes. - [Destin] That's a good practice. Are you gonna be swapping any
hardware here in a minute? - I can. I actually do need to swap that CPU. - [Destin] Do ya? Can I get another camera to do that? - Yeah, sure, yeah. - [Destin] Is that okay? So this is the warehouse. What do you do here? - What do I do here? - [Destin] Well, what does the warehouse? - (laughs) Oh, this is
where everything gets built, where the inventory comes in and we put the computers
together and we test them, we burn them in, stress
test, and then ship them out. - [Destin] Okay, so the CPU noise or whatever you want to call
it I'm hearing right now, those are production computers
that are being tested. - Yeah, those are customer computers that are going through a test right now. - [Destin] Okay, cool. So the one that I speced out with William, is that being built? - It is, being built right now. - [Destin] Like by one of these people? - Uh huh. - Let's do it. So this is the actual box
I speced out with William being built right now. - Yeah. - [Destin] Okay and so
this is the housing. Nice little dolly you got going on here. And so it's just like Legos, right? - Yeah, yeah. I mean, it's like Legos but things can go wrong quite often so we have to build a
complicated process around it, but yeah, yeah. So we have a lot of the
core components preassembled in here, that way we can test them before we get them all into the box and then they'll be assembled. - [Destin] And so once
everything is assembled here, at that point in time would
you run those benchmark tests that you were talking about? - Yeah, yeah. So then once it's assembled,
that takes about three hours. Once it's assembled, then we can start on our suite of tests. We start on basic level tests
to test the hardware function at a very raw level and then once we get the
operating system installed, then we'll actually run rendering tests and Photoshop tests. - [Destin] I'm looking right here. This is the hardware that I
was specing out with William. So that comes to the technician
that's building it here and then he just assembles it all. So do you have like a big
bank of components somewhere and you just go get the
right things off the shelf? - Yeah, yeah. - [Destin] Can I see the Lego box? Basically where you keep all the stuff? - Yeah. - [Destin] I mean, you have all kinds of different cooling systems. Why do you have so many
different types of fans? - For all the different shapes and sizes of computers that we build. We'll have ones that are this big and we have computers,
actually have computers that are the size of like a
hockey puck at this point. And so we need different
size fans to do it all. - [Destin] Really? That's cool. What else do you have on
the shelves back here? Can we go look or is
this all the cool stuff? - Well, I mean this is a
lot of more the cool stuff. This is all video cards. - [Destin] Oh yeah. Yeah, so you got some of the, okay. You could not acquire
those several months ago. - [John] Yeah, yeah. - [Destin] How'd you do that? - Well again, you have
to have a good partner and a company that when we are in need, we could call up and say
what can you do for us? What's nice again about
not being a huge company is if we say hey, can you bail us out, they're used to people coming to them and saying I need 1,000 video cards. We just need 50 in order
to keep our business going through our shortage period. So it's nice that we're
big enough but not too big. - Wow. I'm taking a systems
engineering course at school and there's something called an ICD, an interface control document, right? And so there's several different ways that different components
interact within a system. You have power that can cross the line, you can have data that'll cross the line, you can have like the coolant for example, so like a material that can cross. And in all these things, heat. So how do you keep up with the fact that some of these processors over here have different cooling needs? But you just had on a
spreadsheet over there, you had all these different components but it didn't see listed out
what the components needed in terms of power or heat transfer, things of that nature. - Yeah, well, 'cause at this stage when we have our work orders put together and we're actually
assembling the computers, that work has already been done. So at that point you can
almost take it as a given that everything that's on that work order is going to work together. But when we're designing the systems, that's when we have to
consider those things. And so we've done, a lot of
is like 18 years of experience of putting together computers and putting them under
a thermal imaging camera and seeing what's necessary
to cool things properly and to get the performance we need. - I just realized that
that's what you are, you're computer systems engineers. Obviously your name says systems but systems engineering
is about integrating all these different components and seeing how they work together. And if you change one thing,
it affects everything else. Are you familiar with the
field of systems engineering? - Well, I mean, I am empathizing with what you're saying right now. I think a lot of people do, to some extent it is like
putting together Legos until you get into the
complicated questions of, well, is it actually
going to perform the job when we're done? Like some examples on that, almost all of modern computer hardware does thermal throttling so you're probably not
gonna break something if you don't cool it adequately but you're not gonna get
full performance out of it. And so it isn't enough just for us to see is this computer stable because it might be stable
but not running at full speed. So that's why it's so important that we actually understand what the computer is gonna be used for and that we benchmark it and we know that it's providing
the performance as expected. - [Destin] Oh, so that's why
the benchmarks last six hours because you have to hit steady-state. - We have to let it all heat up and yeah. - Oh, so it's not just
like processor steady-state and power draw, its thermal steady-state. There are mechanical
concerns like heat transfer. Interesting. - And we have to run the
benchmarks multiple times and we do a kind of a
histogram of the scores to see are we actually
seeing a spread of scores that's bigger than we might expect? If you run the same benchmark
over and over again, we should have a result that's within a certain
percentage range each time. And if it isn't, something's going on. We might have a mechanical
problem of something with moving parts or
we have a heat problem. - [Destin] So where you set
the computer in the room matters for the benchmark. - Yeah, yeah, yeah, good. We have to plot all of our scores against ambient temperature so we have ambient
temperatures in the warehouse. - [Destin] Did I just score points? Did I just get credibility? - Yeah. Because, you know, it's no good to us if it runs great here in December if we're gonna be shipping
it down to Alabama and has to run in the summer in Alabama. - [Destin] In high humidity because the air has
different heat capacity. Do you have a thermal
systems engineer on staff? - Not somebody that was classic, that has like a degree
in thermal engineering but that is a competency
we have to have here. - Interesting. Okay, cool. You're making things click. I'll keep listening now. - So we know like what temperature all the different components
are rated to run at and we also know just from math that the way that cooling
works is we're looking at a delta over the ambient temperature. And so we'll plot
temperatures against ambient but we have to take into account with the ambient temperature is so that even on hot days-- - [Destin] It's the delta T that matters. - Yeah, yeah, yeah, exactly. - [Destin] Right, because
you can't dump heat away as efficiently on a hot day
as you can on a cold day because the difference in temperature. Is that right? - Yeah, yeah. So it can kind of get away from you and so we have to have
enough safety margin that even at a higher ambient temperature, we know we're still in a safety zone. - [Destin] Are you using only? Can we look at your fans real quick? - Yeah. - [Destin] Are they all over here or where they at? - [John] You have fans
on the shelf right here. - [Destin] Okay, where? - Here. Yeah, all in this area. - [Destin] Okay, so your fans. So traditionally these
fans are just blowing air over a coil of some type,
right, or a heatsink. - Yeah, a heatsink. - [Destin] So is that called convective, forced convection cooling? Is that what that's called? - Yeah, I guess so. Yeah, you probably know
better than I do at that point but yes, yes. And it's all about knowing
where to put those fans, what speed they need to run at, and how we're gonna like
from a strategy standpoint, how are you going to get the
heat out of the computer? It's great being a custom computer builder but it also means that you have to know all those kind of fundamentals in order to make sure you're
not designing a computer that has like a heat capacity problem. - [Destin] So are they all liquid cooled or is it mainly just conduction? - Most of 'em are air cooled, yeah, yeah. Sometimes you have to
get into liquid cooling when you're in a tight space
and you need to move heat to another space very efficiently. But when you're making
either smaller computers or you have a lot of space
that you can deal with, then actually air cooling's
a better way to go. - [Destin] Do you ever
use Peltier type stuff? - We have experimented
with that in the past. I can tell you about this fun
project we did at some point. Typically don't have to though. Like you can do-- - [Destin] Do you have any
Peltier, what are they called? There's like-- - Yeah, yeah, TCs, thermoelectric coolers. - [Destin] Okay, I just know the words. Like I don't think I have
an understanding of this. - Oh, okay, yeah. So-- - [Destin] It's basically a thermocouple being run in reverse. - Yeah, you apply a voltage and it creates a temperature
difference across the plates and that makes your heatsink hotter, which is good because now you
have a bigger heat delta T that as you blow the air
over, it's more efficient and then it runs through your processor or whatever you're trying to cool at a much lower temperature. We typically don't have to do that though. That's a pretty extreme measure and it's very inefficient way to cool. - [Destin] Why is it inefficient? - Well, we have to apply a lot
of voltage across the plate. - [Destin] Oh, you're wasting
energy to get rid of heat. Okay, I see. Is there any way to harness that heat? - I thought about that too. I think that the difference in temperature just isn't great enough. Like we could make some really low voltage and maybe, you know, I've
had thoughts in the past of well, maybe we can
actually just run the fan based on the temperature
difference of the heatsink but I think at that
point it's just a cost. Like it's not worth the
cost of implementation. - [Destin] It doesn't work. Interesting. Okay, cool. You took me to a deep
place, that's what I wanted. (laughs) It's good to get out near the hardware and then talk about stuff, that's good. Sweet. What's up with all these
upside down bins here? - Oh, these are racks we
use when we kit orders. And so at this point we
just don't have a lot of orders kitted to go but
they would look like this. - [Destin] Did I just
hear a Mario coin sound? - Uh-huh, uh-huh. - [Destin] What did I
just hear, what was that? - That's the order interface. - [John] That's an order alert. - [Destin] So when
somebody orders something? - Yeah, you'll always hear
that little alert happen and then order received. Or it's a different sound
for order adjustments too. - [Destin] I'm Destin. - Alan. - [Destin] Alan, what are you
working on over here, man? - I'm working on the build for a customer, about to put the CPU in and heatsink. - [Destin] Yeah? Are you doing that right now? - Yeah, I was about to actually. - [Destin] Can I watch? - Sure. So just undo these. - [Destin] Is this your favorite part? - [Alan] Yeah, but you
got to be very careful. - [Destin] Okay. - [Alan] The most important part is that you line up this golden
arrow with that on the board. - [Destin] How many of
these do you do a day? - So I started like very recently so still learning like the ropes and stuff but I think I'm doing
around two at normal pace. So what they want though in
like a normal day is three, so I'm still working up to that but I think I'm getting
a good feel for the R5s, which are sadly going
out of life soon so yeah. - [Destin] So you kind of, you build skills up for certain hardware. - Right, yeah. - [Destin] Yeah, how
long have you been here? - This is my fourth week. - [Destin] Oh yeah, so you're learning pretty quick then, huh? - Yeah, yeah. - [Destin] So what's with
the webcam over here? - Oh, so this was just for, this is like a recent addition. I think John might be able to explain that a little bit better. But basically when we're
in the order interface and we do updates for each
part of the order in status so that the customer, if they want to see a
specific part of the order, they can go to the order interface and look at each update updated
by one of us in the team and this webcam is usually here to take a picture of all the parts. - [Destin] Oh really? - Yeah. - [Destin] So as it's being assembled, people can see what's going on? - Yeah, we do one with all the
boxes before it's assembled and after when the system
is fully assembled. - We really don't want to
be like a faceless factory and so we do a lot to
humanize and keep the, we really want the
relationship with each person that we work with and so yeah, when we can do status updates. It kind of looks like a Twitter feed, like what you see when you log in and see the status updates, and we love to send pictures
along of what's going on. - Oh, it's like oh, here's
your little baby puppy, puppy's growing up. We're gonna wean your computer soon and give it to you, it's gonna be nice. That's awesome. Thanks dude. It's Alan, right? - Yeah. - [Destin] Yeah, pleasure
to meet you, dude. - I'm a big fan of your channel too. - [Destin] Thank you, man. - I recently saw the plasma
slow-motion video you had and I really liked the frames per second that you can see when they both collided and you can see that
coloration mixing and stuff. - [Destin] Oh, you're talking
about the toroidal vortex? - Yeah, yeah, yeah, the vortex. - [Destin] Yeah, it's cool, isn't it? - I recently saw that one. - [Destin] That's awesome, man. Thank you very much. Thanks for showing me what
you're doing there, Al. - Yeah, just a little peek. - Yeah, little peek behind the curtain. It's like visiting The
Wizard of Oz kind of. - Yeah, yeah, yeah, yeah. - [Destin] So are you the wizard? - Oh, I don't know. - [Destin] So these are
the real time order status? - Yeah, yeah. So we had a more generic
sound that would play. A part of the interactivity
is that we announce to everyone when the orders come in and allows us to jump on them right away. And of course we have a lot of gamers here so we had to make it a Mario coin sound. - [Destin] I have a question. What's up with the
10-minute red thing on the, and I saw something over here it said 10. - 10 at 10. - [Destin] What does that mean? - Well, actually it's a
little bit obsolete now. We used to meet once every day to synchronize all of the work for the day and get everyone on the same page. And it was a 10 at 10. So we would meet for 10 minutes and wherever we were at that 10th minute, the meeting was over. - [Destin] Oh yeah? Don't do that anymore? - No. Well, now we do every Friday and I don't know how long it takes. It's probably a little
bit more than 10 minutes but we still, we don't like long meetings. - [Destin] Yeah, I don't either. - Don't think anyone likes long meetings. - [Destin] Sweet, cool. What's going on here? - This is our-- - [Destin] You have a laser engraver. - Yeah, we have a couple. This one we use for
etching logos onto things, which actually-- - [Destin] What the heck, dude? Are you about to do mine? - [John] Yep. - [Destin] Really? - Uh-huh. - [Destin] So you planned all this. - [John] Yes, this was planned. - [Destin] This was planned. Oh, okay, cool. Look at you being all fancy-schmancy. - I put a note on your work order saying to wait until you were down here. - [Destin] Oh really? Okay, cool. I was like what's going on, man? How long does it take to burn? - This one will probably take
about two or three minutes. - [Destin] Really? - For the full logo. For a small logo, it's about 45 seconds. - Really, that's cool. And what was your name? - Aaron. - [Destin] Aaron? So you're about to laser
etch the logo on that? - I am. - [Destin] Do you guys do this often? - Every single computer. - [Destin] Really? - Yep. We put our logo on every one. If a customer requests a specific logo, then we're sometimes able to
accommodate those as well. - Okay, so full disclosure here, I don't remember sending you a logo so I'm assuming that happened. (laughs) - We got one and it was
really great quality. - Okay, all right. Let's see how this works. Alright, let's do it. - We got the laser machine
for making different bracing for inside our computers to
keep everything safe in transit. But once you have a laser machine, you find all these different uses. So now we're making our tool holders and paper towel holders and
all these different things around the company. - [Destin] Paper towel holder? Really? Oh yeah. - [John] It's just like when
you get a 3D printer, right? Then you want to 3D print everything. - [Destin] Okay, so yeah. You paid an engineer quite a bit of money for that paper towel holder there. - Well, it was practice. - [Destin] (laughs) There
you go, that makes sense. So did you load up the file on the laser? - I did, yes. I edited in Adobe Illustrator and got it to the right
size and everything where I wanted it to be on the panel and then loaded up-- - [Destin] So did you kind
of claim the laser engraver as your own? - Oh, it's absolutely mine. - [Destin] Okay, just so everybody knows. - Yes, this one, the laser
cutter, the 3D printer, they're all mine. - [Destin] Okay, go for it. - [Aaron] Alright. (whirring) - [Destin] Oh man. Oh, that's fun. Oh. You like this, don't you? - Oh, it's fun. - [Destin] Yeah. So where do the sides of
the computers come from? - Those are the front panels so they actually come on the computers. We're able to take them
off, it's only two screws. - [Destin] So you pop them off
and then you customize them? How many have you messed up? - A lot. - [Destin] Really? - Very, very many. - [Destin] What do you
do when you mess up? - We get a new one. A replacement isn't too hard to get. We have tons of spare parts so replacing a front door's pretty normal. - [Destin] Really? - Yep. - [Destin] That's awesome. - Sometimes they come to us scratched, sometimes they scratch
during the building process, and that goes with plenty of parts. We have lots of replacements on hand. - [Destin] And so that goes back to whoever sent the parts to you, right? - Yeah, it kind of just goes in the trash if it's too scratched up. - [Destin] Got it. Wow, look at that. That's cool. A little moment of vanity here. - Right? (laughs) - So you try to understand your customers and how much your
customers love themselves and so you put their stuff
on their own computer. I see what you're doing there. - It's great when the
computer can get your job done but who's not excited
to see their own name, their own logo on their own computer. - [Destin] Yeah. So it's little stuff like that, right? Got it. That's cool. - I really enjoyed, I watched
you were getting the amps to pick up that little card. I imagine that took forever to make. - [Destin] Took quite a while. Yeah, that's pretty neat. So I don't know. Peeing on a piece of paper
in the middle of the jungle and then you have an ant's carrying it, that's pretty strange thing to do. So yeah, yeah, good point. That's fun, man. So does it work differently
on different colors, like a black case'll etch better? - There are some we can't etch. We only edge three doors,
four doors right now for four different cases. The silver version of the
R5 etches wonderfully, the black version doesn't work out so we don't touch those. - [Destin] So you have to
adjust the focal length of the laser or? - This one actually auto adjusts. The laser cutter that we have in the back, that one you have to adjust by yourself. - [Destin] How does it auto adjust? - When it first came out, it kind of did a, it went down and then popped back up a little bit and that's it measuring the (mumbles). - [Destin] Really? - Yep. - It's more depending on the material we have to change the power, right. You want to get through
just that top layer but you need to get all the
way through the top layer and you don't want to go
past the second layer, so it depends on what it's made out of. - [Destin] What? That's crazy. - So like it's pretty
common in our industry to have anodized aluminum where it's just that very out-- - [Destin] That's pretty cool. - It's really fun. I've etched my own personal stuff onto my stuff as well. I have a R5 case at home and it has an etching on it. - [Destin] Does it really? - Yeah, it's just fun. - [Destin] That's awesome. So where'd you learn Illustrator? Did you learn it because of this? - I did, yes. And for the most part, I only know what I need to
know in Illustrator for this. That's one of the tools
that I use very sparingly. - [Destin] Really? - Yeah. (mumbling) (beeps) - [Destin] How cool. - There it is. - [Destin] That's pretty cool, man. Wow, so what happened to the burnt paint for the anodization? Where'd it go? - For the most part, it all got pulled out into the filter down below. - [Destin] The ventilator? - [Aaron] Yeah. Quite a bit of wind going
through that at any moment. - [Destin] That was a good investment. - Yeah, yeah. - Make people excited about the computers. - Well, yeah. I mean, like what we
originally got it for, I can show you over here. What we originally got
our laser machines for, was for bracing the components. So we're pretty enthusiastic
about making quiet computers and to make quiet computers, you use very large heatsinks and then you ship these
very large heatsinks and they just, they can
cause a lot of problem. And so-- - [Destin] What do you mean? - Well, just they turn into
little wrecking balls, right? You're trying to ship
really big heavy things that aren't supported very
well inside the computer. A lot of people-- - [Destin] And they vibrate in transport. - Yeah, yeah, and a lot of
people what they'll do is they'll use expanding
foam inside the computer. We don't particularly like that. It's a step that customers have to do, you know, they have to
remember to take out the foam and it's shocking how many
people don't take out the foam. And so now we make braces out of acrylic. - [Destin] That's what this is right here. - [John] And it's, I mean, it is solid. It is totally taking the place. - [Destin] So the fans, the fans could support their own mass but you just make that just to support it. - It's not enough, yeah. And especially as we're shipping
things around the country, it's just not enough
secure so we make this. We have the same down for
the video cards down below and it takes it from being
something that would sag to something that's super secure. - [Destin] So walk me through the air flow in a system like this. I'm seeing the fan down here. Is that pulling from the bottom? - Yeah, that's pulling from the bottom and we have a filter
that goes underneath that so we're not kind of
sucking dust off the floor. - [Destin] Is this just
like an off-the-shelf case? - It is an off-the-shelf case. What we'll end up doing in a lot of cases, so this particular one doesn't have that, but we'll end up using
acrylic to move fans to where we need them and
we may take out a fan, we may add a fan. For the most part, most
computers have more fans than they need because
there's fans running that are cooling things
that don't need it. And so us, it's about
like intelligent design of that process. We want to look at things
under a thermal imaging camera, find out where the heat is, come up with a strategy
of the most efficient way to get that heat out of the system, and then we put the fans
where they need to be to get that job done. - [Destin] So, oh wow. So it's not normal to have two fans per. - No, in this case we needed to have a little bit more airflow. This is a little unusual of
a computer for us actually. To have multiple sockets. We had to experiment with,
well, would it be better to have the fans going
up, going out the back? So right now our overall flow
is we have air that comes in, some of it gets sucked into the
video card and out the back. The rest of it continues along to here, gets pushed through both,
and then out the back. - [Destin] So it's
going in that direction. - It is feeding from one into the other and you have to weigh, so this one's getting
hotter air than this one and that's one of the
things we have to weigh when we come up with a design. And in this case, that
was a better way to do it. - [Destin] Is that a dual CPU system? - It's a dual CPU system, yeah. - [Destin] Really? - This is a 48 core system. - [Destin] What the heck is this? - This one's used for scientific work. - [Destin] Where? You can't tell me, can you? - I can't tell you that. - [Destin] Oh wow. Do you build government stuff? - We do build government stuff, yeah. - [Destin] Like what? You can't tell me that either, can you? - Well, I mean, we do
something for the Army, for a lot of government
contractors, for NASA. One that we can talk about,
oh, a really cool one. We do VR simulation boxes for a contractor that services NASA and
astronauts doing training-- - At Johnson Space Center. I've been to that lab. - Have you? - [Destin] Yeah. - Those are our boxes. - [Destin] Those are yours. - For the VR training. Yeah, and we got to go do that too and that was a lot of fun. - [Destin] Yeah, I think it's neat how they simulate the
mass of a spacecraft. Did you see that? - I wasn't one of the
people that got to go. - [Destin] Okay, gotcha. That's cool. So you're touching all
kinds of industries. Really, that's awesome, man. Sweet. - [John] That's cool, so you
got to see it, that's awesome. - [Destin] Yeah, it's really neat. They're cool dudes. Me and a couple of them, we
follow each other on Twitter and they know things. - I know, I was jealous
because they went down there, Chris and Eric were the
two people here that went, and they were giving them the tour and they said hey, do you want
to go into the control room? So they let them into the
control room, awesome. - [Destin] Heck yes. (laughs) Do you wanna go see where
astronauts are controlled from? Yes. That's awesome. - For sure. - [Destin] Is it cool if I get an update? - [John] Yeah, sure. Looks a lot more like a computer. - [Destin] Motherboard's in there. Put you to work, man. Want to see your... So why is this sectioned off here? - Because cutting
acrylic makes more fumes. And it's not toxic but it's not-- - [Destin] Unpleasant? - Yeah, exactly. - [Destin] So are we allowed to go in? - [John] Yep, absolutely. - [Destin] You're cutting right now? - [John] I believe we are. - [Destin] Oh yeah, smell it. Way slower, isn't it? - [John] It is but it's
going through three acrylic. - [Destin] Look at those
interesting patterns it's making. - [John] That's flames
shooting out the bottom. - [Destin] Those are flames? - [John] Mm-hm, because
there's a paper backing on the bottom. - [Destin] Oh, I see what you're saying. - And then there's a
honeycomb underneath that so you're seeing-- - [Destin] The honeycomb
is underneath that so that, yeah, I'm with ya. So it can support the acrylic. Sweet. Can I set a camera up in time-lapse? - Oh, yeah, sure. - [Destin] Look at that. - [John] That's basically
what we're making right now. - [Destin] That's it? - [John] Something like that, yeah. - [Destin] What is it? - [John] That is the support
for the video card bracing. - [Destin] So dude, what's the trade-off between like melt time of the laser and how do you do it? Like can you increase
the power of the laser and do it faster but it
melts too much on top versus the bottom? - Generally we try to keep
our settings pretty exact. We work with three different sizes. So I guess you could. You could go a little bit
faster and a bit more power but for our thickest acrylic, 3/8, we actually do max
power that that'll allow which is 80 watts and
that's still fairly slow. - [Destin] Really? - Yeah. That's what you saw
going in there right now. That's about as fast as we can
get going with 3/8 acrylic. - [Destin] Nice. - If I remember right, like we actually want to cut at max power because the longer you keep
the laser in a particular spot, the more all the surrounding
material gets hot and that's how you get fires. - [Destin] It's about managing heat. - Warping and inconsistent sizes. - [Destin] I just noticed that behind you. What's the battle dome over here doing? What is? - All the monitors? - [Destin] Yeah, what is this? - So this is where we
image the operating systems onto all the computers. And so we can do 32 computers at a time and we have a monitor for each of them. We'll put the operating system on, we'll install the software. This is where all that, we said every computer gets, we run Photoshop tests on, like that type of thing,
that happens here. And so one of the things we have to do is have a monitor up on every computer because if it starts giving
graphical artifacting, we need to see it. - [Destin] Really? Because that would be a function of the graphics card, right? - Like when you have a
problem with a render. - [Destin] Right, like I had in the earlier video I was rendering. - Right, and it might not
be apparent in the benchmark and so we have to have eyeballs on it. - [Destin] So a human in the loop. Eight 12, interesting. And so, oh wow, look at all this. So is this like this is a station here. And so will this be known as station 50 or something like, whatever, station 10? - [John] They're numbered somehow. This is station two. - [Destin] Station two. And so you can just plop a
computer tower right here and then it'll map to one of
those monitors over there. - Right, and then through the network, we'll install the operating system. It actually only takes a couple minutes to install the operating system. The rest of the time spent here is setting up all the software, all the settings in the operating system, the user accounts, like
all of that type of thing. And the stress testing, the burn in. - [Destin] So the stress
test, how long does that take? Six hours, you said? - It depends. It depends on the application. - [Destin] You called it the burn in? - Burn in, yeah. You want to burn in but
not burn out the hardware. So you want to stress
it to an adequate level that we are confident that it's going to perform the function out in the field. How long it takes depends
on the application. Like a system for photo editing is a much less intense burn in process than something that's used
for like machine learning. And so we have a process for
every one of those use cases and so if you buy a computer and you say I'm using it
for Photoshop and Premiere and rendering and machine learning, that's gonna take a long time because we have a lot of tests to run. - [Destin] Oh, I see, I see. So you can make computers
that multitask like that. They might not be super dialed in for one particular application but you still have to run
all the tests for everything. - Right. - [Destin] Gotcha. - Right on. - [Destin] Sweet. I'm gonna set that up real quick. It's the last one? - [Man] Yeah, that's the last one. - [Destin] That's fine. Are you gonna pull it out? - [Man] Yep. I'm not trying to hover (mumbles). - [Destin] No, you're good. We're just gonna leave it like that. Just do your thing, man. (bangs) That is a lot of plexiglass. Little warm in here? - Yeah. - [Destin] Oh, it is, isn't it? Hey, I'm Destin. - Lauren. - [Destin] Lauren? Nice to meet you, man. - [Lauren] Nice to meet you. - [Destin] So what do you do here? - Right now I'm doing quality control so I inspect each system
that we build, all of them, and ensure that they meet our standards for build quality and also
in operation within the OS. - [Destin] Really? - Yeah. - [Destin] So I see a
FLIR camera right here. What's going on with that? - So this system here is
running a stress test, so I'll put it under a
load so we get as much CPU and if you have video card,
video load as possible. - [Destin] Is that the
stress test right there? - Yeah, this is the
video stress test itself. There's actually a CP one
that's going behind it, it's not showing. But it's just to get the
system essentially as warm as I can get it and we're
just trying to make sure that we're seeing adequate
air ventilation through it to make sure that it is
cooling itself properly. - [Destin] So do you have thermocouples like on the inlet and outlet? - No. - [Destin] Yeah? How are you doing that? How are you making sure
it cools itself properly? - We have a temperature range, so there's a temperature graph. - [Destin] Oh, oh yeah,
the built-in thermocouples. - Yeah, the built in software for it. - [Destin] Oh, for the FLIR,
so you're doing it on the FLIR. - Yeah, I mean, all the major components, the motherboard, the CPU, the video card, they all have their own sensors built on and we log those sensors. What this will tell us is one, do we have a rogue chip somewhere that's running really hot that isn't supposed to
be running really hot? And then two is our overall strategy on cooling the system effective? - [Destin] Got it. - If we open this up and like
everything is bright white, then we haven't done it right. - [Destin] So you actually
open the system up. - Oh yeah. - [Destin] Okay, so you have
the side cover on it right now to bake it basically. - Well, not just that, to also make sure we have noticed in QC that there are differences if
you leave the side panel off versus on and so we like to duplicate the customer's experience. So they're not gonna
have the side panel off on their own system so we want to make sure
that we're representing the same airflow that would move through. So like this guy here, if
I let the side panel off, would get incredibly hot compared to if I actually have it on, it allows the air to
move through it properly. - [Destin] Okay, yeah. It's just like a heat,
what are they called, a heat exchanger. You want the air to flow
in a certain direction to dump the heat away from
these coils right here. Is that, do you call this a
coil, what do you call this? - That would be the heatsink. So these are just like the
the fins for the heatsink that allow heat dissipation. So the heat goes from the CPU onto a plate that all these pipes run through and all these little fins are just there like your fins in a radiator. Just attached to it and as
the fans blow through it, they help exhaust the heat out. - [Destin] And the name of
the game is to get cold air on those as much as possible. - Yes. - [Destin] Yes, so let me ask you this. So with your FLIR camera,
we were talking earlier, so you've got two CPUs in
this particular system. So air comes in here and then some of it goes
out with the graphics card but then it comes here and then it goes straight through there. So do you see in the data that this CPU is hotter than that one? - Yes, yes we do. - [Destin] Like imagery, you can see that. - Yeah, and I actually brought this one in so we could show you because this guy has very limited hardware so it's not as easy to see the difference but this one is very apparent. But we will see that. We prefer to have these pointed up so they don't exhaust to each other because this is more of server stuff. - [Destin] There's a vent right there. - Yeah, but they're blocked
off just because for noise and for quality we just
do that on these ones. If we could rotate these, which with these new
processors and the design, you can't really do that. But we would like to see that because this is more server setup when you have a lot of airflow, so you don't get that much of an impact. But in this one, there will be a good sometimes
almost 10 C difference between the two CPUs. Which is the big difference. - [Destin] Really? - Yeah. - [Destin] Yeah, so inlet, just movement, and then this is the outlet here. - Yep, we actually have two inlets. There's also one up front but
it's more for the hard drives. But yeah, so these are
essentially our two inlets that pump all the air in the chassis and then essentially CPU fans, the GPU, and this fan all push it out. - [Destin] Is it possible
to wire the fan up backwards or are they, I mean
connectorized so they can't? - You couldn't do it connection wise. You'd actually physically have to like-- - [Destin] Oh, I see. - And we see that sometimes in QC. Someone, you have a bad day, a terrible Monday or whatever, and you'll accidentally have
like two fans blowing in on each other on the CPU and make some interesting results. - [Destin] No bueno, yeah. Do you catch it with
thermal or do you just, can you look at it? - Usually that's more visual but you can sometimes catch it with just our benchmarks, we'll sometimes see like oh, hey, the thermal on this one's
really high, why is that? And you go look at it again, you can see the fan's
pointed the wrong way. - [Destin] So what's your background? How did you end up here? - So I was actually an auto
mechanic before I started here and I started here actually
building in assembly and I kind of float around more often. I actually work here in
QC, in building computers, or also in the shipping department. - [Destin] Yeah. Well, I mean, auto
mechanic work these days is becoming more and
more, what's the word? I mean further away from
the carburetor, more-- - It's more electronic
and more digital in a way because everything now is run
by the computers and whatnot, so I do have a lot of
sort of digital experience having to do things like oscilloscopes. - [Destin] So are you seeing crossover between your work on
cars and your work here? - Sometimes. - [Destin] I mean, you're troubleshooting and once you find the problem, there's a way to eliminate what, I mean how to isolate the problem. - Yeah, so I guess like in some ways, following a car's wiring diagram is kind of like following
the motherboard manual for oh, if this is the issue, follow to this point or whatnot. - [Destin] Because
finding out the root cause of the problem's the same
no matter if it's, you know, a gasoline-powered mechanical system or something like this. That's interesting. Sweet. Are you about to fire that up? - I was gonna fire this up, yeah, if you wanted to get some video of this. Because this will be the
good one for benchmarks and thermal rather than that guy. - Yeah. So you know what's gonna happen already, you're just doing it to show me. - Yes, yeah. - Okay, cool. I will accept that. - [John] We would have done anyway, right? - Yes, well, I was going to anyway. This is the last part of our
big order that we're doing but we do it on each one to make sure that they are running right because these run hot. - [Destin] So why are you taking that off? The IR can't see through the plexiglass? - Yep. It reflects the IR, so we got
to make sure it comes off. Also when these guys, we've discovered that the motherboard and the CPUs make it so that you can't just use one brace. We have to specially make it
otherwise these get tweaked a little too far and
with these processors, that's not good by the way they're set up. - So is this a true test then, because you're changing,
how's it a true test because you're changing the air flow? - Well, I am slightly but not too much. We don't have a side fan, so really all this would
do here is just block off, which actually in a way
this slightly makes it just a little bit warmer but not too much. Maybe about one C. - [Destin] With this
side off, you're saying. - With the side off, yeah. - [Destin] Yeah, with
the side on, it's cooler. So this is kind of like a worst-case. - In actuality, this
actually helps a little bit because it allows all the
air just go straight through instead of allowing some
of it to come around here and like loop back around. So we don't want a dead spot
of hot air being built up and then suddenly going through, so this kind of helps the CPUs only push. - [Destin] Got it. So basically by taking that cover off, you're gonna have hotter
CPUs than you would if it was in its assembled state, so we're gonna see a hotter computer than you normally would. - No, not so much actually. - [Destin] Really? - Yeah, no. It would be like maybe one C hotter but you won't really see that
much with a thermal camera. It's just on there for bracing. That itself won't affect
the temperature too much. - [Destin] Sweet, cool. So what's happening? - So I started up the stress test. This is one of our more
specialized systems, we can't use a normal stress test. So I actually open two programs. - [Destin] What's going on here? - Yeah, so this one right
here is running the CPU and it's opening up essentially
a worker per thread. So if it's hyper threaded
and you know it's a 10 core that's actually 20 threads, there are two CPUs in here
so it actually has a total of 48 total cores with 96 threads. - [Destin] You're saying
a bunch of letters and numbers and stuff,
so what is a thread? - So a thread is essentially, I guess it'd be a working
instance for the CPU. And when you think of threads, if it's hyper threaded, it will have two. One physical and then one that is I guess sort of an existing one that's a copy of the one. So these are 24 core processors so that means each
processor has 48 threads and that's a lot. These are very big, high end processors. - Yeah, I was just sitting here thinking about how many that was. It's a lot. - Yeah. I built a couple before that have been some very specialty things that we've done and
like almost 200 threads. - [Destin] Really? - Yeah. So it takes a while, took about a minute just to load all of them. That's a long time, usually-- - [Destin] So when it was opening up, it was opening different threads. - It was opening a thread, yep, each one. And then behind it because on these guys, to properly load the RAM
and the video card up, we actually have to use two softwares. I have another one back here
that's just running the GPU so it doesn't have a nice graphic of like the donut moving around, it's just a completely digital only. - [Destin] Oh, I can see
it warming up over there. - [Lauren] Yeah, it's
starting to get warm. - [Destin] Is that the
graphics card warming up first? - [Lauren] Yep, you're
seeing the graphics card and the RAM are all starting to get hot. The graphics card will usually
get the most hot first-- - [Destin] Do you mind
pointing out where the RAM is? Yeah, you can you can
walk straight across. - [Lauren] So essentially
here's the graphics card and it's, you know, they're
getting hot like normal, and these are all the RAM chips. And the higher capacity RAM chips get hot. - [Destin] And how much RAM is in there? - 380 gigs because this is
that new six channel set up. - [Destin] So it doesn't
look like there's air that's flowing to the RAM. - That is kind of correct, yes. Not as much air flow gets to the RAM in normal standard desktops. - [Destin] Why? - Mainly just because
not that much air flow actually moves through the chassis. In a normal server setup, there are fans moving, they're very loud, and they push a lot of air so those tend to get hot. But even in like-- - [Destin] Are these fans on? - Mm-hm. We like quiet computers. - [Destin] The fans are on. - Yeah, yes. At load. - [Destin] At load. - And it'll get a little bit louder but the fans that are
gonna get the loudest will be the CPU fans because they're actually going to change, the case fans themselves will
stay one speed the whole time. - [Destin] Really? Okay, so can you please talk me through what you're seeing on the screen there 'cause I mean, these are your
diagnostics tools, right? This is what you do. - So what we're seeing
here is that the video card is actually running properly because we're actually seeing it get warm and it's getting hot, as
much as I would expect. It might get more to
the yellow, orange-ish. And I'm also seeing that the memory is actually starting to load properly. So I am seeing that our test is running essentially perfectly at this point because I'm seeing all
the sticks change color. Sometimes when there's an issue, you might only see one half of it and then I know that, oh, for some reason the other CPU or not all the RAM is being utilized. And we want to make sure
everything's being used on the stress test. - [Destin] Really? Do you ever see one, so
a dead chip is one thing that could happen. Do you ever have one that's inefficient and it gets too hot? - Sometimes. Some of the very high
capacity, so 64 gig modules, if you don't cool those properly, they will get incredibly hot. Like enough to burn you if you touched it for just a second or two. You'd actually have a
second-degree burn from that. - [Destin] Wow. - Yeah, it's just the higher capacity, they just get incredibly, incredibly warm and we want to make sure
especially in those ones that they don't get too hot
because if RAM gets too hot it can actually fail and be bad. - [Destin] Like unsolder
itself kind of stuff? - No. - [Destin] What's the
failure that you see? - It'll start causing kind
of memory corruption issues in a way. It'll start like dropping or starting to act finicky. That's the extent that I've really seen. Not to the point of
unsoldering itself, no. Normally most stuff, it'll
either outright just die before it would unsolder itself. - [Destin] That's cool. Do you like your job? - Oh, I love it, it's great. - [Destin] Yeah? - Yep. - [Destin] Not leaving? Where'd he go, where'd he go? He's right there. - Not that we're planning on. - Yep, not that we're planning on. - [Destin] (laughs) I just
like to make it weird. Just I like to see John
squirm a little bit. - I learned something here actually. So I mean I always had this view that this thermal imaging was more about and it is about making sure
we're cooling things properly but I thought that was a great point that I had actually never considered that actually the absence of heat is, it's just as revealing as too much heat. - Especially with these ones too because this is the most problem system where I've noticed that because if the test doesn't work properly, it'll drop one of the CPUs
and stop loading the RAM. And so that's one of the ones where it's really critical to see. So like when we built
the Quad Xeons before when they have a terabyte of RAM, they can take 40 minutes
to load all the RAM but I make sure it-- - [Destin] I have a
question, I have a question. I think you said a terabyte of RAM. - Yes. We have built systems with a terabyte. - [Destin] That's a thing? - That is a thing. - [Destin] That's a thing? - Yeah, in scientific computing mostly. - Yep, we can actually
technically go up to four if we really wanted to but we haven't had anyone
buy one of those yet. - Quite expensive. - Oh, unreasonably expensive. I believe the 64 gig
modules go between seven to $900 a stick and you're gonna need
32 to get to that point? - [Destin] Holy moly. Good gracious. - Yeah, you're starting to come
to the realm of car payment just in RAM only. - [Destin] So is that, I mean you're like unraveling the universe with computers like that? - Computers like that, we'll see-- - [Destin] I'm not being sarcastic. - No, I know. I mean, for machines like that, a lot of times these days we're seeing it for machine learning. Where we end up needing huge
amounts of RAM like that would be like fluid dynamics
or molecular dynamics. We need to load this really
big complicated system into memory so that all the
computations will run on it, it's very efficient. - [Destin] So you have the ability to make these insane computers and you're selling these to people. If I were to own a computer that has one of these
terabytes of RAM, you know, all these different
chips and all this stuff, what do you do with it in the downtime? - Yeah. - [Destin] Like is it
best to just turn it off and not use it or should I dedicate it to
like scientific something while it's not being used? - Yeah, I mean, so I think
that was where I thought you were going with this question that yeah, I mean, for
the machine that valuable, you should not have downtime. So it's about maximizing resources. And a lot of times when we
sell a computer like that, it's going into a university
or a government facility where there's many users so it really essentially
doesn't have downtime. - [Destin] Really? - There's always somebody waiting for it. With scientific research, it's all about there's never
enough hours in the day to get all the computation done. And so when we sell a
new, really high-end, high-performance system, what used to take them
months might take them weeks but it's still weeks. - [Destin] So if it's important to keep these things running constantly, do you also have some form of
endurance test that you run? - Yes. And we want that to be about the specific
application you're running. So sometimes that's really tricky. Like on machine learning
or scientific computing, it's their own code that they're writing in Python or on different
machine learning frameworks and so that can be a challenge for us. But what we'll do is we
want to run stress tests on a computer, we call it stress testing, that is as close to the
real-world application as what we can replicate here. So if you tag a machine, if we know you're using
it for like Photoshop, we will go and we'll run
Photoshop stress tests on the machine. The challenge is burning in hardware but not burning out hardware. Get to a point where you're
running so much stress tests and now you're impacting the
longevity of the hardware. But we do need to run
enough tests that we know that it can maintain this
work load for weeks at a time. - [Destin] So your goal is to perform non-destructive
verification testing. - Yeah. And so where does the
line of destruction begin is one of the questions. - [Destin] Is that where
we're at right now? - [Lauren] The camera reset itself. - [Destin] It recalibrated. - [Lauren] Sometimes, yes. So I'll just recalibrate-- - [Destin] No, that's fine, that's fine. - Yeah so (mumbles). If you leave it like this, this is the automatic view so it sort of just auto regulates between, so we're seeing the highest
it's getting to is 67 degrees. But we have a different scale
range that we like to use just because 67 is not actually that hot. - [Destin] So you like to
look at a different scale so that you consistently
know what you're looking at. Let me ask you this. Do computers cool down in
the same way as they heat up? Meaning like if you were to cut
this computer off right now, would that stay hot for an
extended period of time? - I mean, it'd still be warm
but it will slowly cool down. I mean it will heat up to a point and then it also kind of breathe, it's sort of like breathing, it'll heat up in a certain way and then it also the cooldown
will also be in a certain. So we'll see the cooler
parts cool down faster than the hotter parts. - [Destin] Do the fans
run after you shutdown? - No, usually not. Yeah, so once you turn the system off, it'll actually fully turn off. Yeah, but at that point also you're actually not
generating any more heat so it isn't necessarily a bad thing. - [Destin] One other question. So when you heat up something, it expands. - Yes. - [Destin] When you cool it, it contracts. So if you're turning
your computer on and off, you have expansion and contraction cycles. - Yes. - [Destin] Do you see any
mechanical fatigue associated with the expansion and
contraction due to thermal? - So generally the way
computers are currently built, that is actually not an issue a normal user would ever encounter. If you are someone who does
the overclocking competitions where you're using liquid nitrogen, that is actually a thing that's kind of, people have found the thermal limits where enough of those times will actually cause the CPU to shatter because it just can't physically handle like the temperatures
they're getting it to. - [Destin] Mechanically. Well, it's a discontinuity
and it's internal stress that's because of it. - Yes, yeah. It'll start to fracture. Usually in this application, no. Normally the average user
will not encounter that. There are generally enough
safeties built into point where if your CPU starts
getting hot enough, it'll actually start cutting its workload or its speed to say, so that way it doesn't
actually encounter that. - [Destin] It buffers itself. - Yeah, the CPUs, GPUs will do that. Yeah, they'll buffer themselves. - [Destin] Thank you very much, man. - Yeah, no problem. - [Destin] This was fun, yeah. So what's your, let me see,
what's your favorite component to look at with the thermal? - Usually the video cards
on most of our systems are the only ones that really
actually heat up the most. Some fun ones are certain raid cards. So a card that handles hard
drive setups and whatnot, they get very hot without
even doing anything. So if you do anything on them,
they heat up very quickly and it's kind of fun to watch. But yeah. - [Destin] A raid card? - Yes, so basically-- - [Destin] I didn't realize that a raid array needed
a card to operate. - Some of them do. And usually in like the
high density servers where you have like 60 drives, they actually have dedicated
raid cards or whatnot. But a lot of, for the most part, raid cards usually
themselves are much faster than on board one on a motherboard will be because it's actually
got a controller itself, then it runs just off the PCIE lanes. - [Destin] That's awesome. - And some of them can get very expensive but they're really cool. - [Destin] How's it going? - [Man] Good. - [Destin] Hey, I'm Destin. - [Man] Hey, it's a pleasure to meet you. - [Destin] Yeah, you as well. - [Man] I've watched some of your videos. I absolutely love the high speed stuff. - [Destin] Thank you so much. Well, that's why I'm here. I'm here so you guys
can help me with that. - [Man] Okay. - [Destin] I'm not joking, I'm having problems
processing all my footage on my computers and so that's why I called in the first place. - [Man] I can see it being a challenge. I imagine some of those files are huge. - [Destin] They're gigantic, yeah. - Destin. - [Destin] And what was your name? - Don. - [Destin] Don. - Don (mumbles). - [Destin] Are you the PhD
guy I keep hearing about? - I am the token scientist, yeah. Yes, PhD in theoretical chemistry. - [Destin] Okay, so
what are you doing here? - Well, it's kind of an
interesting question. You know, I moved out to
Seattle with with my wife. I had been doing consulting and I'd been working doing Linux clusters for computational chemistry. But moved out here and
I'm a hardware junkie. So I was doing consulting, I was like yes, I'm not gonna do that anymore. I was looking around
for something fun to do and so I'm looking and you know, I've been doing computer
stuff for a long time. I did lots of science back in the day. - [Destin] Didn't you do something with, the word quantum was in
whatever you're working on. - I'm a theoretical chemist, which mostly means quantum chemistry. And so essentially everything
I do is mathematics. And I just had just a wonderful thing. I did a series of posts on
using a GPU framework Pytorch for general scientific computing and an example I did was
some quantum mechanics code. So I actually coded up a
whole problem and ran it and showed how to optimize
the code for GPU and all that. Was just stunning and
it was a blast to do it. So yeah, so quantum mechanics
has come up recently in stuff that I've been doing. - [Destin] But do you, so as a PhD, do you kind of represent
the scientific customer? - For here? - [Destin] Yes. - Yeah, I'm the scientific adviser here. And this is something
I really, really enjoy because I went from
doing just pure research into actually working with
clusters and stuff like that and code implementation to
facilitate scientific work and that's really, really satisfying. - [Destin] That's interesting. So it's like if I could
draw a loose metaphor here, like if you had a paleontologist who is in the field digging, you are the guy making the
ground-penetrating radar to help him find out where to dig. - There you go. - [Destin] You're kind of
helping produce the tools to enable the scientists. - Exactly. And there's deep satisfaction in that. You know, it's that being able to help your colleagues do
their science, super satisfying. And it keeps me, because I haven't done active research, I haven't published
papers for quite a while, but I have the freedom here to actually do some really fun
stuff, which thank you, John. (laughs) Yes, I do appreciate that. There's like a series of articles, like oh, quantum mechanics,
what are you doing? Yeah but they were (mumbles). So but then it's
interacting with customers and stuff like that. But part of the reason
I'm down here today is helping one of the
consultants with a customer that is having some performance issue and some multi-threaded
code that they're writing. And you know, I've been doing
this stuff since the mid-90s. - [Destin] So that's fun stuff. - It's fun stuff, I love it. And being a hardware junkie, that was why I came down
here in the first place. Right about the time
that I was looking at it, you know, I need to do
something with a good company. I want to just stay doing computer stuff 'cause it was really fun. And the Intel Xeon Phi, or if you're a mathematician
you might call the Xeon Phi. P-H-I. Math is phi. (laughs) But-- - [Destin] I'm on the
engineering side, I'm with ya. - It had just come out and it was like I was going man, I really want to play with that thing. How can I even do that? So I came down here and I
talked to John and Chris and the other guy that
was here at the time. I'm going, you guys interested in doing like more scientific computing and high-performing
computing, stuff like that? You know, maybe looking at
some of this new hardware? It's like yeah, come on down. - [Destin] That's awesome. - And I've known about
Puget Systems from way back. John, I'm sorry John, I'm
gonna mention the fish tank. (laughs) - [John] The fish tank, we haven't talked about the fish tank yet. - When I was working with
parallel quantum solutions, you know, a bunch of PhD
nerds doing computer stuff and fun chemistry. But this is in the old days
when we're doing cluster nodes, cooling is a big issue, everybody is looking
at how to do anything. John and the guys down
here did an oil immersion, a system immersed. - [Destin] Oh, the whole
thing was under water. So they sealed off all the electronics? - It's in a fish tank-- - [Destin] Did you do that? - We did, yeah. - [Destin] So you basically
waterproofed the parts. - We didn't have to though because we put, it was in a fish tank but we
submerged it in mineral oil which is non conductive, not capacitive, you just put your electronics in. - [Destin] Did it work? - It did work. It did work. - [Destin] Really? - Uh huh, and we did it for years. - [Destin] What? - It was brilliant. I mean, it's not a practical solution but now, I mean, you can go to some HPC data centers now and they are actually
doing stuff like that. There's all kinds of
immersion cooling options, some crazy wonderful-- - [Destin] That's like science fiction. - Stuff from 3M and stuff. We don't need that and this was always just
kind of a fun thing to do. But I remembered Puget Systems from, and this was, when was that, John? - [John] Must have been '05 or something. - Yeah, way back, you know. And every computer nerd in
the world knew Puget Systems from doing that, like that is so cool. And we all talked about it and we've thought about
it and stuff like that. So I came down here and I'm looking for people to work with and good companies and that's, I have got to go down and talk to the guys at Puget
and see what they're doing. - [Destin] And they said yes? - And they said yeah. And it's been fantastic because I think we caught it
really at a really good time because we've done really
well with scientific computing and machine learning, all that. I mean, recently that's big and this stuff is right up my alley, so and you know I was in the
perfect position to do this and they were in the perfect position to have me on board to do it. So I'm having fun, everybody benefits. - [Destin] That's awesome. - The company's benefited, the customers definitely benefit, and-- - [Destin] You're getting
to do more science. - I get to help my old
colleagues and stuff like that. - [Destin] Where did you,
did you teach at all? - I did a little bit. - [Destin] Where'd you go to school? - I my got my PhD at
WSU, Washington State, a really great theoretical
chemist there, Shuster. I went from there and
did research in a lab at University of Arizona with (mumbles), who's wonderful and we did
just really crazy stuff. I mean, yeah, this is kind of out there
quantum mechanics stuff that we did. Very much-- - [Destin] What about optical? When are we gonna quit using wires and go to like fibers? - It's already actually
being done a little bit. - [Destin] Yeah? - Yeah. Well, okay, there's a whole lot. (laughs) If you get me going, you could really get me going. - [Destin] Well, just
answer it concisely then. - That question, there's
actually at last Supercomputing, which I need to talk to you about, John, because I really do want to go again. Last Supercomputing, there
was a startup company there and they were using, what
do they call that process? The DLP chips, in projectors? - [Destin] Yeah. - That's a big massive video array. They were actually using that for compute. - [Destin] Really? - Yeah. It was amazingly clever what they did but this was actually really
literally optical computing. And it was very limited
in what they could do but what they could do was blazing fast because they could essentially compute all this stuff at once. - [Destin] Right. So how long till you guys
here at Puget Systems have desktop quantum computers and optical computing methods here? - I think especially on
the quantum computing, you know, the quantum processor stuff, I do know quantum mechanics
and I know the weirdness of it. And I look at what is being done. I have not spent the time, I just have not wanted to
go down that rabbit hole, but I look at it and I have, there's ideas that like totally make sense. Like taking advantage of
things like electron spin, spin states. This is essentially what
they're really kind of doing but you know, they have these
(mumbles) and all the stuff and they had to bring things
down to near absolute zero in order to do the computing. Probably five to 10 years
away from practical solutions but we'll see it. Because they're throwing
huge amounts of brains and money at it. - [Destin] Oh, yeah, so it'll happen. - It'll happen. - [Destin] You just put an X on it? - [Man] Yep, X is the best way for the thermal paste to spread. You can do a dot if
it's a circular heatsink but since these are square, you do an X so that no matter what when
you put the pressure on, it expands kind of out from the center. It's like an extreme version
of if you imagine you put a circle of thermal paste, you'd get like an air bubble
right down in the middle. So by doing an X, there's no way you can
have any air bubbles so you don't even have to be super careful when you apply it. You just have to make sure
you don't have way too much. As long as you don't have way too much, it spreads really nicely. and we swap CPUs, you know,
upwards 20 times a day so we get to see what it looks like after it's been spread many,
many, many, many times. That is definitely the best way to do it. And power switch. Power switch. I already turned it off before. - [Man] You turned it back on. - [Man] Yeah, I turned it back on. (laughs) It didn't turn on, so that's good. - [Destin] So you pointed out this. That's a Mac, right? - Yep, yep, that's the
top of the line Mac Pro and hasn't been updated since 2014, 2013? - [Destin] Which is kind of why I'm here. - Yeah, yeah, exactly and that's why a lot of
our customers come here. So yeah, we have that one. We also bought the, I think it's actually the top
of the line iMac Pro as well, fully kitted out because, yeah, we get a lot of customers
that they are on Mac now but they are thinking about moving to PC and they kind of want to know
like what performance would-- - Well, let me ask you this. So I'm kind of here because I feel like Apple
might be losing their way in that they're no longer listening to the needs of the customers. - I would agree with that. I don't want to like bash on Apple because Apple does awesome things. - [Destin] Okay, so we
don't want to do this. We don't it we don't
wanna talk about that. - No, I think that's fine. - [Man] Just don't wanna bash them. - Okay, so it's like this. I edit on this right here. I try to edit on a MacBook Pro, right? It can't handle 4K at all. And so I bought an old, you
know the old Apple towers where you can actually put stuff in it? - Yeah, yeah. Those are great. You can upgrade 'em. - [Destin] Yeah, I bought one of those and I've been upgrading it but it's no longer doing what I need. Like it can't handle what you showed. It can't handle the high-speed
camera timelines and stuff. - And I think a lot of our
customers are coming to us moving from Mac to PC for a
couple of different reasons and one of them is upgrade ability. You know, you buy one of these new iMacs, I mean they're awesome
pieces of technology. Like the engineering
going into it is amazing. But say in the future you
want to put more RAM in because hey, you were
working with 1080P footage and now you're working with 4K. You can't do that. You want to put in a more
powerful CPU in the future, you can't do that. You want to put in a more
powerful video card, whatever, you can't do that. - [Destin] So what you're
doing in this room, like you've got processors
and you mate that up with certain types of
graphics cards and hardware and then that coupled with
different types of software gives you a certain result. And so the ability to be flexible
and all that is important and I can't do that on Macs now. - No, well, and on Macs,
a lot of the things that we run into a lot is, this is starting to get a little technical but they only use AMD video cards or GPUs. So it's only AMD and that's
a big deal for a lot of stuff because CUDA is in video only and a lot of software
runs way better with CUDA. I mean there's some rendering
software that is only CUDA so you have to use a PC if
you're going to be doing that. And so just to have that freedom with PC, you can use AMD, you can use Nvidea, you can use whatever. Hey, you need FireWire, you
don't need to use a dongle. We can put in a FireWire PCIE
card and you can just go. So it's the customization,
upgrade ability, and just straight-up performance. I mean, if you have a budget of $6,000, you're going to get more
performance out of a PC at that price point. - [Destin] Are you fluent in Mac? - Fluent in Mac? - [Destin] You know
what I'm trying to say. Like do you have the ability to use a Mac as your daily driver if you wanted to? - I could. Well, because we actually
have even articles where we looked at all
these different applications that we target. Premiere, Pro, Photoshop,
Lightroom, After Effects. And we had to recreate all
of our benchmarks on the Mac. So like I even had to get into
Apple scripting and all that. - [Destin] Well, let me ask you this. How do I know as a consumer
that you're not telling me to use a PC with all of
these bells and whistles because of your confirmation bias, because you're most comfortable with it? How do I know that? - Well, some of that, I
mean some of it comes back to the whole transparency. So we can say trust us and you don't have to just
trust us on blind faith. - [Destin] You have data
is what you're saying. - Yeah, we have data and
we help people with things. So I'll even bring up... Let's see, that's an
article I'm working on. Let's see, like in here. So on our Premiere Pro page, we have a whole tab that is
are you moving from Mac to PC? And we try to address
here's the common questions that people ask, what are the
hurdles people come up with? Especially in video, ProRes
comes up all the time because you can't export
to ProRes on Windows. So like how do people get around that, what's the way to fix that? And by just presenting all of those, like here are the
questions you probably have and here's the answer. And again like we were
saying kind of earlier-- - [Destin] But my
question from a consumer, because you're more intelligent
than I am in this area. It's just a fact. - Yeah, it's just experience. - [Destin] Yeah, exactly. So in this particular area, how do I know that you're not
trying to actively push me towards these solutions? How do I know that you're
giving Apple versus PC a fair shake here? - Well, I mean, the easiest way is to
talk about performance. I mean, all the other things, a lot of it is personal opinion. But on the performance thing, again I think it is just we put all of our benchmarks out there and we say here's the test files we used, here's the settings we used, and here's the data we arrived at. We have open comments on our articles so if people really
disagree, they can talk and we can discuss it and maybe, because maybe we do have a little bias. I mean, I'm sure there
is a little bit of that as much as we try to be
as unbiased as possible. I mean, that's just not
possible with human nature. So by leaving that stuff
as open as possible, allowing those conversations, I think that's how we get across that we're not trying to sell
you something that's wrong. - [Destin] So do you run
the same benchmark tests on the Apple products? - Yep, and that's why we have
those machines over there is because we did that. We did the iMac Pro when that came out and then we'll probably
include those again when like Intel does a
major refresh of CPUs. That's not something we do all the time because our benchmarks
aren't as automated, so it takes a lot more work to test Apple. But that's, so we just do
it whenever it makes sense. - [Destin] Got it. But you don't have to do it often because they don't release often. - No, yeah, Apple doesn't release often so it's really only when
there's a major shift on PC that honestly will almost
always make PC look even better because I mean, yeah, like you said, Apple doesn't refresh. - [Destin] Because the
update loop is quicker. So like the tests you're
running right now today, have you run these same similar benchmarks over on the Apple? - Similar. I mean, our benchmarks
are currently evolving as things change and everything like that but yeah, very similar things. And then like when the
new Mac Pro comes out, that there's all the mystery about, when that comes out, we will probably buy one of those and do the whole round. - [Destin] So one phrase
answer, devoid of emotions, a cold hard test engineer answer. If I want to edit 4K 60
frames per second right now, am I buying a Mac or am I
buying a a Puget Systems PC? - PC will get you a lot
more for your money. That's probably the most
straight way you can say it. You can edit 4K on a Mac. You're gonna pay more to be able to do it. - [Destin] Okay, got it. So you can do it with, they
prefer Final Cut obviously. - Yeah, obviously. - [Destin] Yeah, so but you
can also run Premiere on there but is Final Cut more
efficient or effective, do you know? - That's kind of the areas where we stay out of as much, simply because there's so much when you're comparing software packages. It's not just like you have a clip, you load it up and hit play, does it play? It's like all of your workflow stuff and it's all of your like effects. - [Destin] It's a more holistic question. - Yeah, and we usually tend to feel that that's outside of
our ballpark a little bit. So we do a lot less telling people what software they should
use rather than they tell us what software they use and we tell them what to
get for what they're doing. - [Destin] Got it. So this is your software package, this is your problem
you're trying to solve, we're gonna help you
optimize for that solution. - For that, yeah, rather than trying to push people into like a different software package. We might, sometimes we'll
make recommendations. So like if someone comes to us and says that they typically record with, especially in Premiere Pro, like they might say they're
doing 4K with MXF wrappers. Sometimes Premiere doesn't
like MXF for whatever reason and so we might tell them,
hey, we can totally do that but are you aware that if
you swapped over to MP4, you might get better performance
and nothing else changes. So like sometimes we'll
give small advice like that but generally no, we're not gonna say like you're on Premiere now, you should really move to DaVinci Resolve. That's not usually what we do, no. - Just get robes and wizard
hats and be done with this. Wizard hat. Just put a wizard hat on top of your head, wear robes in here, and like I don't know, have a dragon statue or something. - You missed Don earlier. He's the PhD-- - [Destin] Oh, no, no, no, I met Don. - Oh, yeah, yeah, yeah. He can get into fun conversations for us but for other people it's
probably super boring. - [Destin] He said the
word quantum quite a bit. - Yeah, he likes quantum
mechanics and stuff, yeah. - [Destin] That's awesome. Well, thank you for your time, dude. - Absolutely. - [Destin] This was a blast. William, thanks for helping
me design a computer, man. - My pleasure. - [Destin] It was a blast. And thanks, I enjoyed it. - Anytime. - [Destin] Yeah. Like literally any time? - Couple times a year. - I was about to say, man. Yeah, you need to think about that. Alright, cool. Well, yeah, I'm gonna go home now and see how the computer works. Okay, John, I have a question. So all these are processors, correct? - [John] Yes. - [Destin] And so what's going on here is he's testing the processor capabilities with different software
packages, et cetera. So my question is back in the day, there was a thing called Moore's law where a processor speed
would double every 18 months. And I mean, like I started out on a 386 SX and just slowly started working up. But it seems like at some point in time, everything changed because
it went from processor speed to power optimization. Is that true? - Yeah, yeah, yeah. Well, we kind of hit the limits, right? And back when I started this company, Moore's law was the big deal. In order to make your workloads faster, you just make a faster processor and you do that by putting more chips into the same amount of area. And it was really around 2005
when we saw dual core CPUs that everything started to change. - [Destin] First of all,
what is a core mean in a CPU? - It's the compute unit
that can execute tasks within your operating system. And so back before 2005,
it was all single core. You had one thread or one
operation that could be happening at any given time and in order
to make your computer faster, you just make that chip faster. But in around 2005, we were getting to the point
where Moore's law was starting to taper off and we had to find other ways to make computers faster. - [Destin] And to be clear, the reason Moore's law
was tapering off was because of the amount
of basically transistors you could physically
put in a certain area. Is that why it tapered off? - Yeah, I mean Moore's law
was just an observation of a trend that was measuring how quickly we were making
technology smaller and smaller. And as you make technology smaller, it also uses less power
so you can use that power to instead make it faster, right? - [Destin] Why did we hit the limit? - Well, physics. I mean, back in those days, we were at 90 nanometers transistor size, went to 45 and then to
22 and we're at 14 today, trying to get down to 10. But it's a curve that's
slowing down over time and as it slows down, there's other ways you
can achieve performance. You can do more cores, you
can have GPU acceleration versus CPU acceleration and have more advanced
instruction sets on processors. So it's a more complicated
way to get performance. - [Destin] Well, let me ask you this. For example you have a GPU
graphics processor unit and a CPU, which is a-- - Central processing unit. - [Destin] Central processing unit. So you have those two things. How do they work with each other? Do they complement? Do you offload tasks to the GPU? I don't really understand that. - Yeah, well, and it's
kind of similar, yes, yes. You have the CPU that
takes on certain tasks that the operating system gives it and the GPU takes on other tasks as well. And they're very different
types of product. The processor generally has very few cores but they're very fast and a GPU has hundreds
and hundreds of cores and they all work in parallel together. So very serial tasks, the CPU is good for. Very parallel tasks, the GPU is good for. - [Destin] Oh, so that's
why you use parallel tasks in a GPU for graphics
because you have pixels on the screen and you
can compute multiple, you know, you can break that
math problem up basically. - Yeah, when one math
problem isn't contingent on another math problem, then they can become very parallel. And so now we're more under
this new thing, Amdahl's law and Amdahl-- - [Destin] What, spell that? - Amdahl. A-M-D-A-H-L. 95% sure that's right. Amdahl's law, that talks about okay, if you have multiple
cores, multiple processors whether you're on your CPU or your GPU, whatever it is, that all
have to work together, it's a measurement of the
efficiency of your software in being able to do that. So think like you have
an engine in your car and you had a second engine in your car. It's gonna be faster, right? You got a third engine, you got a fourth engine,
got a 17th engine. - [Destin] There's a principle
of diminishing returns. - At some point, yeah, it's
gonna be diminishing returns and the infrastructure it
takes to have all those engines can't be overcome by
just having more power. - [Destin] Oh yeah. It's just like the same when you play like one of these rocket games where you can add more rockets but you're pushing more
rocket, yeah, okay. Or it's like a teacher dividing up a task amongst the classroom. At some point, the teacher
just spends all their time making the students do things instead of just helping them
learn how to do it, okay. - So it's more about the code, it's more about the software
than as it is the hardware because the software has
to really efficiently use all those different cores. - [Destin] So what does Amdahl's law say? - Amdahl's law is a measurement of how application can
scale to multiple cores. And if it's 100% scaling, then every time you add a core, you just get more performance. But almost nothing is 100% scaling so we find that as we add
more cores to a system for a specific workload, at some point you keep adding cores and you don't get more performance, you get less performance. - [Destin] So you're telling me that if I buy a computer, I don't necessarily want more cores. - No, no. And that's a very common thing that, and so by extension if you
buy a more expensive computer, there's a point at which you're going to get lower performance again. - So when you're building a computer on one of these websites that just let you go and pick performance and what kind of hard drive
do you want, all that stuff, when you're building the computer, I naturally click on
the most cores possible. So you're saying it's possible I'm shooting myself in
the foot when I do that? - Yes. It depends on your application. Some applications will scale just fine across multiple cores. Other applications, no, you're better off with fewer cores. It's gonna be a more
efficient application. You'll actually run
faster with fewer cores. - [Destin] How can a person know? - Well, that's where we. You have to know software just
as much as you know hardware and you have to do all this testing and you have to figure out
through empirical testing what does the performance
look like as we change, as we turn all these little knobs, what does it do to performance? And then we can know what to recommend. - [Destin] So that's
why you're paying people to sit here and swap a
CPU out on this one device over and over and over with that GPU. - Yeah, try this CPU, try
this CPU, try this CPU. And what we're gonna end
with at the end of all that is this curve where
we'll see we saw better and better performance and then it started getting worse again and we'll know, okay, that's
where we should recommend. That's their sweet spot. - [Destin] So you're looking
for the inflection points. - Yeah, yeah. We're looking for that,
if you have this curve, where is the optimization
point on that curve? - [Destin] Is there a way
you can simulate all this instead of having to physically run it? - Sometimes we can, yeah. We can put in a very high-end processor. Well, once we build that curve, that's where Amdahl's
law kind of comes in. Once we build a curve for
that software package, then we can extrapolate what it will do when we give it more or less cores. But the challenge is that how
quickly everything changes. The software keeps changing,
the hardware keeps changing, and so really there's no
substitute to just testing it. - [Destin] Oh, I see. Because by the time you
invested the talent and money to build the simulation,
it's already changed, the goalpost has already moved. - Nothing stays still. - [Destin] Got it. So it's just easier to just
brute force through the tests and run the-- - Yeah, well, and like
we'll know over time like the the general characteristics of the software don't change over time, so we'll know. We're not gonna like shotgun
approach the testing. We'll know where to start
and where to optimize from. - Got it. That's awesome. Thank you very much. So simply put, what
does Amdahl's law state? Is there a way to boil it down? - Amdahl's law I think is
actually more of a formula. So it, oh, you have it up? - [Matt] Yeah, I brought it up. - [John] We have an article on it. - [Destin] Oh, what does it say? - We wrote an article on it a while ago because we actually used
to do our testing that way. We would use one CPU,
highest score count possible, and we would disable cores and benchmark the same program
over and over and over. And we did that for a while but especially as like the
video card is being used more. And like these different
CPUs have different amounts of these other little features in them like different capture
mounts, different frequencies. We found that it's actually
become not as accurate. You can get very general
idea by doing that but it's just as fast to just throw in all these individual CPUs and just-- - [Destin] Do it. - 20 CPUs and then you have
real firm numbers for each one rather than doing that
kind of scaling stuff. So yeah, we used to do that but at this point, I think it's-- - [Destin] This is more robust. - This is more robust, it's more accurate, and honestly it's probably faster because if we really
want to do that scaling, you have to start at one core. You want to know that math formula, you at least need the result from one core and running some of these
applications on a single core will take days whereas our
benchmark might take an hour. - [Destin] Do you ever have manufacturers question your data? Do they ever come to you and say we're better
than you think we are, obviously you did something wrong? - Not usually, no. We've had the opposite sometimes. The Lightroom team, they came to us and said we know these
things weren't optimized. We made this new version of Lightroom called Lightroom Classic and now we're gonna be doing this stuff, make sure you test it
and publish the results. - [Destin] Oh really. - So sometimes it's actually
kind of the opposite. They let us know like hey, we
made all these improvements, check it out. - [Destin] That's interesting because Lightroom's running behind you. - Yeah, right there, yeah. - (laughs) That's pretty cool. Man, this is hardcore. How do I edit this video? There's so much information here. There's so much information. Anyway, I'll figure it out. That's good. - Just keep deleting video
files until you're down to like four and then
just edit it from there. - Yeah, there we go, there we go, that's how I'll do it. - [Matt] That's why we're
not video editors here. - [Destin] Man, it's 4K too. I need to... Once the camera came
out, everybody got quiet. What's up with that, man? - [Man] That's a good question. - [Destin] So do you guys
personally believe this? Is that the deal? - Personally believe what? - [Destin] This thing that's
on the wall behind you. - Oh, well, I mean we
have tattoos of it, so-- - [Destin] (laughs) You
have tattoos of that? - Live and breathe it. No, absolutely, this is great. - [Destin] Yeah, that's pretty cool. How long you guys worked here? - I've been here for just
under a year and a half. Ben here, Broden just started. - [Destin] Broden? - Yeah, Broden. About two weeks ago. - [Man] Two, three weeks ago. - [Destin] How's it going? - Oh, it's going great, lovin' it. - [Destin] Yeah, you dig it? - Oh yeah. This is fun. - [Destin] You guys handling
orders, what are you doing? - Oh, we work in the support department, so I'm the support lead. My name's Ben. And yeah, we handle the
inquiries that come in. - [Destin] (laughs) Is
he making fun of you? You making fun of him over here? (laughs) - Yeah, we handle the inquiries, so when there are issues out on the field, they call in and we'll take care of them. - [Destin] That's cool. You get a lot of calls? - It depends on the day, it
depends on the Windows update. - [Destin] Oh really? - That week, but yeah. - [Destin] Is that pretty much it? I mean every time there's an update? - And there are hardware
issues and software issues, so. - Really? Cool. Thanks for your time, guys. Appreciate it. Real quick, true story here. When I was in fourth grade,
we used to have this program with a local supermarket where if we collected enough receipts, they would give you in
your classroom a computer. And in 4th grade, Mr. Keenan
was at the front of the room and he was opening our
box of this new computer and in the back of the room, I had been watching Arsenio
Hall when I was younger and I just thought this is what you did when you were excited
because I remember seeing it when my parents were watching it. So at the back of the room, I
went hoo, hoo, hoo, like that and I still remember Mr.
Keenan turned around. He was a preacher during the weekends but he turned around,
he goes who did that? And it was me. And he said out in the hall, young man, and he gave me a paddling. I got a paddling for being
so excited about a computer and I remember as he paddled, he said I feel bad I have to do this because this is one of my best students. So all that to say computers
are really interesting when they're being unboxed and I mean like clearly
that's a formative moment in my childhood but I realized that my kids
haven't had that experience because it's always been
laptops for them and iPads and things in their life or tablets, whatever it is. So when the Puget
Systems computer came in, I wanted them to be there to help me unbox and set up the computer because it's a full-on desktop machine. I don't know, it's a workstation and they've never seen that. So it was really, really fun
watch my son's face light up when he powers this thing up. It's just a really neat moment, something that reminded
me of my childhood. I don't know why but
it made me feel things. You've never seen a computer
be opened, have you? - [Boy] No sir. - [Destin] Check this out. - Why is it in such a big
box if it's just a computer? - How big is the computer? - [Destin] Because it's
like the real deal. You've never seen-- - How big is it? - It's in that box, we gotta open it. Do you wanna help? So when I was a little boy, you had to spec out everything
you wanted in your computer and it can make it fast or slow depending on whether you
got, that's not sharp enough. And so this, I used to get
your super, super excited when Pawpaw would open a new computer but I haven't done that. Have you ever seen a new
computer being opened? - No sir. - Have you? - No sir. - [Girl] Take off the top of the box, flip the box over gently. - And then pick it up, what's it say? - [Girl] Pull it up. - [Destin] Oh, is it gonna fall? - [Boy] Got it. - This is the bottom. - This is the bottom. - Oh, look at that. - This says take everything off. - You put this in the transmogrifier? - [Boy] Oh, (mumbles). - [Destin] You know what that looks like? - [Boy] What? - That looks like sleep. - [Boy] So does this mean
you're gonna get to eat dinner with us a lot more? - [Destin] Yes. - [Boy] Go in like this. - Yeah. - [Destin] You have the honor. Who's gonna do it? - [Boy] Do what? - [Destin] Who's gonna hit the button? - [Boy] What button? (beeps) - [Destin] Is it running? They said it's super quiet. Does it sound like it's going? - Fan's on back here. - [Destin] Huh? - Fan's on back here. - [Destin] I can't even hear it, can you? - No. - [Destin] Can you hear it? - No sir. - It might be on caps lock. - [Destin] That's the license agreement. You're supposed to read that entire thing. - Is it gonna take a long time? - [Destin] Go ahead and click accept. - Good, I don't have to read it. - Hi to you too. - [Boy] We're getting
everything ready for you, Dad. Might take several minutes. - Should we go eat? Okay, I hope you enjoyed this
episode of Smarter Every Day here on the second channel. I want to say thanks to Puget Systems for letting me go check out the plant and all that good stuff. The reason I got to them
in the first place is I was looking for a specific
hardware application for my software needs. So they have a website
that details all that for very different builds
like a Twitch streamer or whatever it is. That's down on a link in
the video description. Feel free to check that out. I guess that's it. Thanks for watching this whole video. That's pretty awesome. Feel free to subscribe
to the second channel if you're into that sort of thing. If not, no big deal. I'm Destin, getting smarter every day. Have a good one. Bye.
