NVIDIA Tesla M40 - A Gaming and Transcoding DEAL...with a catch

breaking news the gpu market sucks right now okay obviously that's not breaking news however there are gpus if you look hard enough that are actually immune to the scalpers and bots that's where this guy comes in and what this is is an nvidia tesla m40 this is a gpu from 2015 based on the maxwell architecture it's got 12 gigabytes of gddr5 ram 3072 cuda cores and it's the same chip that's in the titan x so a pretty solid but old gpu and the best part is that you can get it for about 170 ish dollars u.s so why is it immune to the scalpers why don't people want this why is it so cheap well the main reason uh it sucks at mining you're not gonna get any use out of it as a miner so that's why scalpers don't want it secondly if you look closely uh let me know if you noticed something strange about this card yeah there's no video output so strange right don't worry we can get around this and thirdly uh there's no fan on it so what's that about so yeah the tesla m40 is not designed for gamers it's not really designed for consumers it having no video output or fans is pretty much an obvious reason that it's not for consumers this is designed to go in a server for deep learning or other graphic based applications and specifically a 1u server that has a lot of static air pressure going from front to back so that's how this will be cooled but we have a way around that and i've been tricking you guys the whole time this is actually empty there's no card on it i've taken it off and it is now looking like this so how are we gonna use the card let's find out [Music] all right so this is essentially exactly what it looks like it is an aio designed for a graphics card i got it off amazon it's from a company called id cooling this is the id cooling ice flow 240 it's a 240 millimeter red connected to your gpu with another fan on it to help cool the vrms and i did not have high expectations for this but it works it works surprisingly well if you want to see how i put this together i did a live stream where i did it all live and if you want to check that out i will link it in description as well as right here it's pretty decently long so feel free to skip through basically it wasn't a bad experience at all getting it put together but there are some nuances obviously when you're working with a very strange gpu like this the front bracket doesn't really have anywhere to mount to anymore so i had to use an m.2 screw with a nut to help it stay secure and it kind of works okay so we solved our cooling problem but what about the fact that there's still no video output well i asked you guys in a youtube poll if you had a tesla m40 what would you use it for whether that's a cloud gaming server transcoding whether that's for handbrake proxies or for plex or something or if you want to do machine learning and the top response was that you would trade it for a handful of cheez-its and a grape kool-aid so i'll be right back hey can i have some ice cheese no what if i give you this nvidia tesla m40 what am i supposed to do with that i'm going to put in your home server okay well can i have the drink no it's mine go finish your video all right she actually let me uh have the grape kool-aid all right so the actual plan with the card is to install on my server and use it as a gaming machine and i also wanted to try transcoding on it and turn it into a kind of automated proxy machine using handbrake where i can drop all my footage from my videos into a directory and handbrake will automatically pick them up and create proxies using the tesla m40 and for machine learning i'm not a machine learning person so based on the specs provided i know it's pretty decent but anybody out there who has experience running machine learning programs or doing any type of work machine learning uh let me know your thoughts on this card but let's go install it in the server and i will be back with some performance updates oh one thing to note before is that the power pin on the back that is not pcie standard that is eight pin cpu so if you have an extra cpu eight pin uh coming from your power supply you can use that for this or if you don't just get a pci six pin to cpu eight pin adapter off amazon for like five bucks so i'm gonna go install this i'm going to install this be right back all right we are back and there are two ways you can go about getting video out from the tesla m40 one way is easier i'll talk about that basically if you have integrated graphics on your processor or another maybe smaller graphics card in your machine already you can just plug the tesla m40 into another pci slot use the output from another source like your integrated graphics and then go into whatever program you want to run on the tesla m40 so probably a game and set that to use the m40 as its graphics it's a pretty straightforward process i will link a article in the description below if you want to go that route i'm going a different route so i'm going with one that's a bit more complicated and it's essentially straight from jeff over at craft computing he did a video a while ago using the same tesla m40 where he wanted to get it set up and running games i am going with exactly his setup so if you want to see how to set this up i will link his video in the description below as well as the guide that he created it's pretty easy to follow i mean you need three pieces of software you need parsec you need tight vnc and you need a virtual audio cable and once you have all that set up you essentially just connect to the machine using parsec and it'll act just like a normal computer and if you're using this across a lan it's extremely responsive we're talking like a less than a millisecond latency now i've done a video on parsec i'm not going to cover it in this video if you want to check that out i will link a link i'll put a link up here but let's dive into it let's take a look at the performance in gaming on the tesla m40 all right so here you can see in parsec my tesla m40 machine is showing up all we have to do is click connect just like that we are into our tesla m40 desktop i'm going to open up cpu id and take a look at the temperatures and the specs all right so here's our tesla m40 idling at 27.28 degrees so by this point some of you are probably thinking oh i've seen shrouds like 3d printed shrouds you can put on the test on 40 to help cool it and those don't work and the reason i know that is because i tried uh i don't know why this exists jeff did it in his video it didn't work i thought maybe if i under folded the m40 i could get some decent performance out of this thing with a little 40 millimeter fan no it sucks so with our 240 millimeter aio we are looking at 28 degrees idle which is cool but the real thing is what about under load stress test let's run it and when i did this with the tiny little 40 millimeter fan shroud thing it instantly over the course of about maybe a minute got up to 90 degrees and thermally throttled but in all my tests this hasn't gotten over 50 degrees that's extremely impressive for that hundred dollar aio so we're not going to sit here and watch this for 10 20 minutes but you can see the gpu is at 100 its power is at 100 this is a 250 watt tdp card so the fact that this hasn't shot up to around 60 degrees already is very impressive so for gaming i ran three games i started with league of legends which is probably the easiest game in the world to run so 1080p max settings we were getting 120 fps which is way more than you need this is limited to 60 fps on parsec so 120 is way more than we can even use and the latency across lan on parsec was only like one to two milliseconds which is extremely playable for a fast twitch game like league of legends so let's step it up to something a lot harder to run let's do the witcher 3. the witcher 3 1080p ultra settings the only thing i turned off was hair works because it's pointless and with those settings we were getting 63 fps which again with our 60fps threshold extremely playable imagine just setting up whatever computer you want in your house and having a centralized gaming machine that you can just pick up and start your games from with extremely good quality wherever you are in the house from any machine it's awesome now to push it further we went with borderlands 3 i set it to high settings and we did dip below 60fps we were sitting around the high 40s sometimes creeping up to 60 in certain environments but again it's playable if you want to step down the quality to low or medium or even the resolution down to 720p you will get a much better experience so gaming on the tesla m40 is actually really good if you want to compare it to an actual g-force card it sits right about where the gtx 1070 is so if you have the prices for a gtx 1070 right now significantly higher than a tesla m40 even with the 100 aio so if you're willing to put in a little bit of elbow grease the tesla 40 isn't a bad buy for a remote gaming server now what about something a data center card like the tesla m40 is more designed for and transcoding my idea like i said before was to throw this at a vm have it run handbrake and automatically create proxies of my videos so that's what i did so in handbrake i took my pie for a week challenge video that's about 35 minutes long encoded in h.265 and 1080p and attempted to make a proxy down to 720p and i did this in h.264 now how did it fare so here you can see the results from my two tests one where i did gpu encoding and one where i did cpu encoding both in h.264 so for a 34 minute video the gpu encoding took about nine minutes and five seconds and produced a file that is 942 megabytes kind of high for a 720p proxy moving over to the cpu encode now this is eight threads of my amd epic 7302 and that took 12 minutes in 30 seconds and produced a file that's 361 megabytes that's less than half of the size from the tesla m40 render so while you do get faster encoding on the tesla m40 you do get a much better compression and quality out of cpu encoding so let's compare that to my main machine over here which is running a ryzen 9 59 50 x and a rtx 3080 ti so gpu encoding h.265 and vanc with that same video doing gpu encoding on the 3080 ti it rendered in three minutes and 44 seconds and took up 806 megabytes so there's your comparison i mean three minutes and 44 seconds versus nine minutes and five seconds yeah but i mean this is a much more expensive card and cpu encoding followed that same trend it finished in four minutes and 58 seconds with a total file size of 240 megabytes so if you're looking to set up an automated proxy encoding server like i am i'd still recommend going with cpu encoding although it does take a little bit longer than gpu encoding the file sizes are much smaller and you do get better quality however if you're looking at real-time transcoding for something like plex the m40 might be something you'd like to investigate you can save a good bit of cpu cores for other virtual machines or services in your home server while just allocating a single tesla m40 to do all the transcoding for plex according to nvidia it can handle four 4k 30 fps transcodes at one time so if that's in your ballpark then maybe this is for you alright so overall what do i think about the tesla m40 well if we weren't living in a world where gpu prices were insane then i would say absolutely not do not buy it however the reality is the gpu market sucks and i can't change that i don't have enough subscribers so yeah help me out subscribe subscribe please it's all things considered if you want to get your hands dirty and pick up a tesla m40 as well as some cooling solution then i mean it's usable gaming comparatively to a gtx 1070 and pretty decent encoding or transcoding for plex if you can snag a tesla m40 and a cooling solution for around 250 bucks i'd say it's worth it otherwise just pick up a ryzen 5700g and call it a day so now i'm curious after seeing this video do you plan on picking up an m40 or if you have one what are you using it for i'd actually be curious to know let me know down in the comments but that's all i have for you today if you like this video be sure to drop a like below if you like content like this please consider subscribing thank you so much for watching and i will see you in the next one [Music] you

Channel: Raid Owl

Views: 1,403

Keywords:

Id: 4NDcXFPB8mM

Length: 15min 4sec (904 seconds)

Published: Fri Dec 10 2021