- This is the second fastest server we've ever gotten our hands on. It's for scientific applications
and machine learning and it might actually
be better in some ways than the fastest one,
which happens to be the petabyte of flash project over there. Why? Because this one is
freaking water cooled. It's got six GPUs in it. It comes in two flipping
and wooden crates. That's really all I know about it because it's been stuck at
the border for over a month due to some kind of stupid COVID delay. So, we're gonna open this puppy up and show ya exactly
what's going on inside. Whoa. - [Alex] Also it's only four GPUs. - [Linus] Oh, I thought it was six. - [Alex] Yeah, sorry. - That's still pretty good.
- [Alex] Yeah. - Pretty good, just like our sponsor, Signal RGB with signal RGB
you can control and sync your favorite RGB
devices all from one app. Best of all, it's free. So, download signal RGB at the link in the video description. (upbeat music) These are wooden crates
sealed up with porks screws. Alex has his way but I think
I'm gonna go try my way. - [Alex] No, don't use that. - [Linus] I can't find a crowbar. - [Alex] There's no way
he's getting the crowbar before I'm done. - Oh, well that was no fun. (Alex laughing) - [Alex] What is in here? Why do we have two boxes? - I don't even under... Yeah, I was gonna say,
I don't even understand why there are two boxes? - [Alex] Are these are GPUs? - This ain't no ordinary
GPU ladies and gentlemen. - [Alex] No. - Now there's a big difference between the water cooling that
we as gaming enthusiasts use and what you'll see in
workstation or server or data center applications. - [Alex] I think this is
all custom by Comino look, look at this back memory cooler. - [Linus] What boards are these? They have NVLink fingers,
so, they must be 3090 class. - I think they're all A6000s. Yeah, I've been trying to
figure out the pricing on this. I don't know yet, it's
all just like, you know, if you have to ask you
can't afford it kinda stuff. - I didn't think gigabyte was
a Quadro board partner though. Check out these quick connects, love it. So your cold comes in here. Presumably does some stuff
heats up and comes out here. - [Alex] It does look like
they're five grand a pop US. That's like MSRP, they're currently going
for like nine grand. - If we screw up the packaging enough we might not be able to ship it back. Oh wow, it really did not
make it through there. Thanks, Jerry rig everything box cutter. You can tell it's serious when it just does not look performance at all. 512 gigabytes of DDR4 3200, six A6000 GPUs, two Xeon platinum 8368Qs,
two SFX-L power supplies. What a reopening? - [Alex] What the heck is this? - [Linus] It looks like meat. - [Alex] Oh. - [Linus] Oh my God,
it's just power cables. Well, that was... - [Alex] Wait, but how many
power cables do we need? - I'm a little confused because there are multiple
power supplies listed on the bomb. Wait, what? What is it? - Oh, that's not power cables. - Looks like some kinda handle. I think I gave you the light side. Oh wow, that's heavy, yep. - It's designed to be rack
mounted, it comes with rails but I can feel little
nubbins on the bottom, I think they're rubber feet. I guess you can kinda use it however. I mean that would be a big advantage to going liquid cooling like
this with your compute node is that you wouldn't have
to put it in a server room. You could actually, ow, have it on your desk and
without being super disruptive. I mean, yeah, that machine out
there might be more powerful but it's obnoxious to be in the same building
as let alone room. - [Alex] Look at this. - Wait, this side's just rad. - I am really confused and
you're about to be confused too. Let the camera see your confusion. Whoop, everything's fine, it's just a very durable LTT store.com water bottle. This particular one belonging to Alex. - [Alex] Well, but yeah, what? - [Linus] It's got GPUs in it already. - [Alex] So what are those - [Linus] They're GPUs smart guy. - [Alex] Oh, well, obviously they're GPUs. - [Linus] It has three power supplies. Shout out Comino for
their custom IO sticker that's actually a really smart
way of branding your machine. - [Alex] What would we use this for? - [Linus] Almost doesn't matter. Okay, well Kyle has that thing upcoming. He's got that raccoon that's
like pooing in his yard and he wants to train like
a machine learning model to recognize it and like shoot it with a
airsoft gun or something. Apparently it takes like four days to do the training on
his like consumer GPU. This you can fit so much
machine learning in here. What's the front Alex? - [Alex] I think this is the front. Wait, but is it? - No.
- Yeah, yeah, this must be the front. - This has to be the front 'cause it's got a thermal
sensor in front of the radiator. So, you'd wanna know
your intake temps, right? Oh wow, are these ever tight? That stripped one could be a problem, we could end up drilling this out Alex. - Oh, I really don't want to do that. - No, it's dead. And it mangled your torque spit. - Oh, great. - One power supply, ah, ah, ah. - Wait, why are you doing that? - I'm powering it up. - But I'm about to drill it. - I mean, that's fine. - What about we don't do that
until we know that I did it. - Whoa, here it goes. - I guess it's on now. What? - It just had a little
panic moment and shut off. - [Alex] Cooling system standby. - [Linus] Wait, is it even full of water? - I can see right here
there's a reservoir. The water's right up at
the top. I think it's fine. Should I just drill this
and get the top off? - Yeah. Look at that now we're
both being productive. Do you wanna get footage of him? Wait, stop Alex. - [Alex] Too late. - Brandon footage.
Don't... Just pretend, pretend to have
the breakthrough moment again. - [Alex] All right. - Hold both buttons to turn on/off. Cooling system start, fan
RPM monitoring on here. It feels like it's intended to
be up like this a little bit. - [Alex] Yeah, kind of. - Yeah, whoa, lots of bubbles. I mean surely if they picture it like this and the text is this way, it's
intended to be run this way. (Linus screaming) - [Alex] There we go. - Good thing I didn't stab that, that would've been pretty bad. Wow. - [Alex] Holy crap. Look at that back plate. - What are the GPUs that are in here? What was the point of
those other ones then? These have HDMI ports on them. So, I do not think that these are A6000s. - [Alex] Are those 3090TIs? - They could be 3090TIs but I don't understand why they have these crazy eight pin connector
adapter doodads here then. Wait, I don't think I've actually
powered on the system, Alex. - [Alex] Like I don't have
any lights on the keyboard. - This controller system
up here is trippy. Here's our temp sensor for
the front, hanging off of it. And then you can see the 24
pin connector goes into here. Oh crap, is that USB? Yeah, I don't wanna
wrench that off too hard. And then you can see fan leads, power, presumably there's a
cooling flow rate monitor. All that appears to be completely custom. Should we take a closer look
here before we fire it back up? - Yeah. - [Linus] These guys have come a long way since the last time we checked
out one of their systems. I recognize this triple power
supply setup from last time that's super cool. We've got three 140 millimeter fans. So, these are taking advantage of the full four U height of the chassis and then pushing that air through what looks like a triple
120 millimeter radiator which I honestly gotta say
feels a little optimistic in terms of cooling. We've got dual Xeons and then quad GPUs that are all supposed to be cooled by this single thick radiator. I love this distribution block here for every heat generating device, you can see we've got
the cold and hot sides coming into this Delrin distribution plate and then we've got cold
coming out to the components. These are all completely
custom water blocks and then hot coming back
to the distribution plate all joining together and
going back to the radiator which then goes to our pump
reservoir combo unit up here. What we don't know is what
the heck these things are. There's two pumps in here by the way. I'm not sure where they
are, but that's pretty. Oh, there they are, I
think they're DDC variance. On/Off commands not available. Just rip out the power just there. Boom, they're gone. - Another news I checked in, we actually have a terabyte of ram. - 64 gigs each. This single dim has more
ram than your workstation. - [Brandon] I wonder how many
people are buying this thing? - I think it's less about how many people are buying this thing and more about how many of
them some people are buying. Is this on or what? Select the Input. - Like I don't even have keyboard lights. - [Linus] Interesting. - Well, first of all let's talk about what went wrong yesterday? - Oh yeah, let's do that. Did I screw it up? - Yes, you did. - Oh, was it when I pulled this thing off? - [Alex] It was, yeah. - [Linus] Cool. - [Alex] So it turns out
that the power switch is like in here. So, this starts up the
cooling checks everything. Make sure it's good. - [Linus] Yeah. - [Alex] And then...
- Powers on the board. - [Alex] Yeah, powers on the board. - So it just didn't. - Yeah, because when you did this, the power switch connector, you can see a little jumper right in there where you like really reefed on it. - Cool. - Yeah, that was unplugged. - Okay. - Took a call with the people in Austria. Jake, you spent what like
two, three hours on this? - [Jake] 20 minutes. - [Alex] Oh, only 20 minutes, not too bad. - [Jake] What did you just do? It literally just blue screened. - 20 minutes... It what? It blue screened? - [Jake] Yeah, yeah.
- [Linus] Shut up. I didn't do anything. (Alex laughing) - [Jake] You just have an aura. - No, I don't have an aura. - [Jake] K, It's working now. - Yeah. - [Jake] Don't, just don't touch anything. - I'm not touching it. - [Jake] Stop, stop. - I'm screwing in a VGA ca... - [Jake] Look users use keyboards, just go to the keyboards, okay? - [Linus] Do I get to know
what GPUs they are now? They have the A6000s in there which means those must be 3090TIs then. - [Alex] They're actually 3090s, I talked to them and they sent them just so we could show them getting swapped and show how easy it is. - [Linus] Okay, cool. What are we gonna do now? Because my understanding
is you've spent some time working on a benchmark that
can actually take advantage of all of this GPU compute. How much memory do these
fricking things have? 48 gigs each? We have 200
gigabytes of video memory. - Like just the CPU screen even like. - 76 cores, a 152 threads, I don't know Intel's
lineup very well anymore. So what? These are each 38 cores. - Yeah.
- Are, is that good? Are they fast? - I don't know, should we test it? - Okay, so we're going up against what, what do we have here
in the reference ones? What ways it compared to a 32 core 2990WX? - I think favorably. - Thank you for that. I wanna feel the heat though. - Oh, just watch this for a second. - Okay, all right, I'll watch. Okay, it was only necessary
to watch for a sec. No meaningful heat buildup
was going to occur. All right then. - [Alex] Wow, that's fast. - [Linus] 71,000, what the hell is in 8368? They turboed a 3.7, base 2.6, 270 watt TDP and a maximum memory size
of six terabytes each. - During cinebench they reached a maximum
temperature of 55 degrees. - [Linus] They also
weren't running very long. Hey, what happened to our power meters? - [Alex] They're right there. - [Linus] We only have two. - [Alex] We also have the sketchy one. - [Linus] Let's use it one more
time before we throw it out. Cinebench, here we go. I can pretty much guarantee you that the eight pin CPU power connectors of each of these two power supplies are going to the two eight
pin inputs on the motherboard. So, we're sitting at
about 850 watts right now. Shall we do something to the GPUs? - Might as well. - Okay, so V-ray cuda. - [Alex] Sure. - Or RTX? - [Alex] Now, let's do cuda, I think that will hit it harder. - [Linus] Here it comes. - [Alex] Oh my God, holy,
how is it not crashing? - It's pulling over 2000 watts. We have to be on two separate circuits. - We, we are, yeah. - [Linus] I wanna feel the heat. - [Alex] Yeah, we're
just 2000 watts from... I'm gonna get a sun pan
boys, this is crazy. We're already at the point
where the incoming air is 22 degrees and the outgoing air is 44 (Alex laughing) which is pretty nuts. And what's crazier is
that if you check this out this V-ray benchmark only
runs for one minute at a time. So, we're barely even
stressing these GPUs, I have to manually restart
it every single time which means I think it's time for us to switch over to
your solid works benchmark. - Okay, I'm concerned that
it's not really going to work. So, I've been having
a problem in the past. It just like renders a bunch of passes. On this they take such little time that I've been having trouble getting them all to get fully loaded. - To its credit, when I'm
not unplugging power supplies it appears to be rock
freaking solid stable. And in terms of balancing the load here let's hit it with V-ray again, no single one of our power supplies is doing an outsized amount of the work when you're hitting the entire system. I think the most is around 850 watts. And given that these are 1000
watt units from Silverstone, that's not bad. I do have concerns about
those poor power supplies drawing in like, no, it
wouldn't even be 40 degree air because they're on this side. - Yeah. - Of the radiator. So actually they're getting
pretty decently cool air. - This is weird 'cause if you put your hand behind the
radiator, it's super hot. But if you put it down
where the power supplies is, it's still nice and cool. - Solidwork's crashed, maybe some kind of driver
issue, we're not sure but blenders opening. - [Alex] I'm curious a lot
of my laptop benchmarks take about one and a half
minutes on really fast ones to like three minutes on like something that's
a little bit slower. I'm kinda curious what this will do. - [Linus] I'm thinking
a lot less than that. Holy (beep) is it done? No, okay, wait is it... (Alex laughing) Did we just witness a sub eight
second blender BMW render? Samples 3000, sure. - Maybe just increase the
number of light paths as well. - Yeah, total lots. - Okay,
(Alex laughing) it still did it in 16 seconds. This is ridiculous. - That was 3000 samples and
10 times as many light paths. - Okay, can we do like
10,000 samples though - [Alex] Is it even
hitting all of our GPUs? - It's not even hitting all the GPUs. (Alex laughing) Is it possible that's misreporting? - Oh, they are all hot, it
probably is just misreporting. - I think it's misreporting because it just did that in 31.8 seconds. (Alex laughing) - It's gonna be a problem
trying to stress this. - Classroom, well, you saw it here
first ladies and gentlemen that is a 15 second classroom
benchmark, 14.66 seconds. - I normally expect that to
be like 10 to 15 minutes. Who would've thought
that the render machine is incredible at rendering. - From 300 to 10,000 samples. It's still chewing off 20
seconds at a time of the CTA. It's definitely using our cores, as for our GPUs we are
at each one of these is pulling about 280 watts. We've been added a minute. We still have about two
minutes and 15 seconds left can confirm we've managed to
fully load down the system. 860 watts on that power supply. 625 there, so, that's just shy of 1500, about 2000 watts total. - Yeah, 2100, maybe a peak. - Oh wow, it's really hot now that the coolant seated up. Oh, that radiator is hot to the touch. I'm amazed at how reasonably
quiet it is though. - Like you could have this in your room. - Yeah. - And aside from your room being a billion degrees in like 20 minutes, you could live with it. And like GPU's been hitting
this for a little while but they're all at like 65 degrees max while drawing 300 watts. And that's on top of the CPU. Each of which also are doing 300 watts. - Oh, it's done, it finished in three
minutes and eight seconds. It looks better though. - [Alex] Well, I'd imagine
there's 10,000 samples. - Well, look, I've never looked
at it rendered like that. Wait, I think you're right. - Yeah, I am right. - They're NVLink, I think
they'll run in four way SLI. - [Alex] They will. - Is that even supported? That's not even supported on the 3090TI. - I don't know. - I don't think so. Now, rendering device, wait,
did you already click run? - [Alex] I haven't clicked run. - [Linus] Okay, so there's only one which would seem to suggest, it's gonna hit all four of them. Here we go. Here's where we find out, if
it's running in four way SLI. Oh, I'm fidgeting. - For some context, I spent a day and a chiller, and a almost killing a 3090 to get a score of 12,000 in this. - 500, 450, 6... holy (beep)
it's hitting them all. So, we're sitting at about 1700 watts that's a little lower than
what we saw in blender but we know, okay? That the CPU is not gonna be hit too hard during the GPU por... Is this running at 120 frames a second? - Yep. - It's up to a 160 now. What we know is that this will not stress
our CPU nearly as much as that blender render was. So, that 300 wat discrepancy is just one of our CPUs sitting ideal. These GPU's are going full grunt. - I guess we should probably calculate about how much it would cost. - 7,500 times 2, there's your CPUs. Okay, use a calculator I give up. - [Jake] About 60 grand. - 60 grand? Don't forget that they just
ship you random RTX 3090s. - [Jake] Did it just crash? - Oh, I think it did crash. - Oh. - It crashed. Oh, she's done bud. In fairness to Comino four
way SLI has not been like properly supported by
Nvidia in a very long time and it's probably a driver issue. Since we're happy with the A6000s in there and we're not gonna swap in the RTX 3090s, I might as well just explain
why they included these. It was so that we could show
you guys just how easy it is to swap out the cards
from your workstation. Comino will provide
these pre-done up cards. So, basically it's just a one screw, undoing these quick disconnects,
slotting in the new GPU, I mean, I guess you should also
remove your power connectors at some point, and then these actually come
pre-filled with coolant. So, theoretically there's
almost no liquid loss when you go from the
old card to the new card and you should almost never
have to top up the coolant which I thought was pretty cool. Attempt number two is
running a lot slower, power usage is lower. It's around 1300 watts. - [Alex] Oh, it's crashing - Oh, and it's crashed. - I don't know if we're getting 3D mark on this system. - I don't know if we're getting away without telling the
viewers about our sponsor. - Ting mobile. Do you like saving money? Ting mobile is a low cost carrier with rates to help you do just that. Almost every phone on the market
will work with Ting mobile and they have the perfect
plan for everybody, no matter what your needs are. Start with unlimited talk
and text for 10 bucks a month or data plans from $15 a month. Their set 12 plan with 12 gigs of data is only 35 bucks a month. And if you need it, unlimited
data plans are $45 a month. You can even share your
data on a family plan and save even more. Ting mobile also offers, pay per use plans with their flex plans charging
just five bucks per gigabyte. Even with those savings, you
still get nationwide coverage and award-winning support. Consumer reports named Ting mobile their number one carrier in America which is a very large place. So check them out at linus.ting.com and receive a $25 credit. - And if you guys are looking
for another video to watch, how about a system that's
even more expensive you can check out this thing over here.
