Nah miners don't want them. Most only care about newer than Maxwell cards. Maxwell and older are prone to what's called dag thrashing or memory thrashing when attempting to mine a dagger hashimoto/ Ethereum algorithm. But those of us who want to play games or need a GPU to speed up whatever work we are doing, this is a great option.
@@sergeykoss not a great amount, for the power use. maxwell has severe deficits in architecture vs newer cards that make it pretty slow for mining even with 12 gigs of memory.
NO. Last time this happened was with the p106 cards and what happened when someone pointed out you can game on them? Nvidia blocked the feature and with the lack of driver support for games, you can't play anything. We need to keep this on the down low so the people that know can enjoy it. And if larger channels cover it, prices will skyrocket from 150$ to 400$, it's not even worth it. It needs to be an inside secret. Once nvidia catches on because more people report on it, all support is gone, graphics mode is gone, drivers are gone, no gaming. A expensive blender card. You don't need performance reports or builds. Build your own, do your own testing. It's a titan x. It performs the same as a 1070. That's all you need to know for game performance. Don't bring too much attention to it before you ruin it.
i managed to get Tesla K20Xm to work with GTX Titan bios, enabling full OC support and making it work in Windows 10 without any modifications to the OS itself. Matter of modifying device id straps in Titan bios so that it gets detected as a Quadro GPU, thus enabling raster rendering by default. And I used Morpheus II cooler on it, fits nicely.
Awesome stuff Wendell. Glad you mentioned the X300 Deskmini again, I picked one of those up shortly after your video on it and paired it with a 4650g and I have to say it has been a wonderful Home Theater/Living room PC.
On cooling these, the Kraken G12 with a compatible AIO will handle that without a problem, especially if you use a dual-120mm AIO (e.g. the Kraken X53). I have two Tesla M40s that I swap back and forth between Folding@Home and some BOINC projects. The AIOs keep those processors cool without any issues. The memory will run hot, but directed airflow can help with that. At minimum, redo the thermal paste on the GPU as it's likely completely hardened over.
@@Arnau975 any tech person on YT with a viewership in thousands can do that. The dude from Craft Computing doubled the prices of the Sun F80 SSDs on ebay the same way.
I ended up getting 4 K80s to pair with an x99 board with a 5960X CPU on it, after tweaking a lot of BIOS settings. Asus x99 Deluxe II I think? It does run them at pcie2 for bandwidth since the K80 are dual GPU cards, but I'm hoping to get cooling working on it all soon and have a decent rendering/computational box for cheap
With servers there are usually really loud fans. Just ask anyone who has ever powered one on and they might tell you they are NOT silent. It is for the simple fact that the components of a server don't have their own fans for a number of reasons.
Nice video man! I'm glad to see a tech TH-camr showing options instead of just showing how great the new cards are that you can't even find to buy anywhere.
I have been suggesting to people to get the K80 which can be had for 200-300 depending on where you live. It is close enough to a modern GPU that you can do decent gaming. You should also be able to use multiple k80s in a system.
Why would you do that. The k80 is just 2 k40s on one card. You can't use sli properly either and you can't use all 24 gb because with sli, vram does not stack. Even with sli, it only performs the same if not worse than a m40, a gpu that you can get for 150$. The k80 performs like 2 1060s in sli which only sometimes matches the m40 in performance. Not to mention kepler has lost driver support, uses more power, has no reflex support, and is generally worse. It's not a good deal even for 200$ Also, multiple k80s? The k80 is a dual gpu card, you can't sli a gpu already internally in sli.
G'day from Sydney. Cool idea. Market forces in Australia - population 25 million - unfortunately make this cost prohibitive (had a glance at secondary market, yikes) to try as a lark.
Was looking at these Tesla gpus for octane rendering a few weeks ago as the cuda count is still pretty good and that 12gb would be nice for heavy scenes.Had read they used to be and are still used in a few render farms. A few bios issues meant I couldn't / wouldn't be able to get them to work which you mentioned. Thanks for noting and testing that.
I remember that I was kicked out from a tech forum many years back for claiming that you could in theory could game on a Tesla if youre able to hack the software. Guess I was right after all
This reminds me of Lucid's Virtu. They used to use an abstraction layer to intercept and redirect api calls to either discreet or inbuilt GPU. Though it's target was to get quick sync to work even when discrete graphics is attached, as intel used to disable igpu when a discrete card was attached. Funny part is when benchmarking . Dirt series (Dirt 2 back then) performed really poorly on that too.
I've been using this with a dual gpu card where the first GPU (which has the display outputs) doesn't work anymore. Now it outputs the frames from the second gpu through the Intel HD graphics.
@@unitedfools3493 No, I bought this Radeon Pro Duo as defective, but I managed to still use it for something with this workaround. That something being a gaming PC for the living room TV.
Good on ya for keeping it at least somewhat functional instead of letting it become waste! Any use you can get out of it is better than it going to the trash.
My GTX 960 although only having 2 GB of memory on it was not expensive and it still runs everything I ever wanted. I am not one of those guys looking for insanely high frame rates but more or less just playing my games at good frame rates.
Last December, i bought a tesla P4 which cost for 60 usd. Regular price was around 92 USD. I got it cheap because the online shop that i've is celebrating a Christmas to new year's eve sale. I am really happy with it. Paired it up with an Lenovo SFF i7-7700 and can play any games at 1080p with low and some with meduim. I can even play God of war III in RPCS3 at 720p. It dips down below 60 fps, even 30 fps but really playable
Could you do a follow Up Video for Tesla M40 Cards? In the forums there is evidence of them working the same way as you describe for the K80s and since M40 basically is a GTX TITAN X (Maxwell 2.0), I am sure many are interested in using these cards, too... Great video, btw :)
I just put together a Frankenstein contraption a month ago. It's a 2017 Alienware Aurora R6 with an intel i7 7700k, a GTX 1080fe and a Quadro K6000. The hurdle was uninstalling all graphics software and installing graphics software from 2017-2018. Once I did that, Windows 10 recognized both cards at the same time. Better still,..Blender recognised and is running on the K6000 card with 12GB Vram rather than the 1080 card with 8GB Vram. How or why that is happening I'll never figure out because from what I have researched,..Blender utilizes the lowest card available rather than the highest,..so??? Now I'm curious what to expect when I rebuild and use the K6000 with the Tesla K80 I found on E-Bay. P.S. I have less than $1,000.00 invested in the K6000 and the K80. I can't buy a RTX 3060 for that. I also upgraded my power supply to a Corsair RM850X 80 Gold and installed 64GB Corsair Vengeance DDR5 RAM. I'm not a gamer so I can't tell anything about gaming performance. I can tell my rendering time in Cycles decreased by a factor of at least three and according to thermals nothing is even getting warm under a heavy load render.
so i have tried the guid and tried flashing a GTX TITAN vbios to my tesla K20Xm card and nothing works... apparently all graphics cards are not equal. can you possibly do a video or series on how to make the tesla card a titan?
have u tryed using a grid card instead of the telsta seen grid cards sit in that 150 range seem like they could be a little better game features wise out of box
Iris Xe on laptops was a bit more impressive than I thought, I mean the same laptop has an RTX 3070 (Secretly Max-Q). The Xe Graphics, when playing the same game, ran at 100% ad gave a very playable performance of 50FPS, on high settings (1080p with SR 1.2) . Lowering the settings probablies would give a more competitive frame rate, and the D-GPU is more power efficient and would give a better performance, even with RTX on medium.
just bought a k20x for 125 CAD plus hipping and tax im at 180 CAD for a GTX titan that Cost 10K plus in its day. im a happy boy .Gonna try all my trick to get it working . Its headless but i can use anything
I can confirm this working on z97 asus mobo and a i5 4590 and a TESLA k40m, followed the instructions and installation, and overclocking worked, MSI AFTERBURNER IS NEEDED. It is noted in the instructions, that you need to run the nvidia-smi commands after every reboot, however if you shutdown you won't need to reenter the commands, with rebooting yes you do. Edit: forgot to add the GPU temps with 3d printed fan included Blower style cooler bought from ebay for about 35 dollars, less than 60c running heaven benchmark, at 1125 mhz core clock, and 3125 mhz memory clock, WITHOUT changing thermal paste or cleaning the card. edit2: overclocked the Tesla k40 to 3154MHZ Memory, and 1200MHZ Core Clock ran Heaven benchmark twice no errors, and temps were under 60c the whole time I am going to clean and replace the thermal paste and any thermal pads on the cards to see if they drop further, NOW the caveat is the blower fan is LOUD but the gaming performance is SOLID!!!
after reading the comments, apparently the M40 was available as low as ~$100 a couple weeks ago, but now its up to ~$300+. M40 would make sense for 980 Ti level performance, but shouldn't a K40 for $100 be more like 780 Ti performance? And of course it would have the pitfalls of Kepler being a bit too long in the tooth, where newer games like Doom Eternal can run like garbage regardless of the decent raw power of the GPU
The nvidia smi and clocks make it benchmark about the same as a 980/Ti. Don't forget you have 12g vram and the 780 2gb wasn't great. This is higher clocked.
It'd be great if you could post some benchmark numbers to see how they compare to a 980ti, I'm really thinking about trying it only because I like to experiment with stuff. Great video!
It's literally a 980ti. The m40. It's the same dye, it's actually closer to a titan x, which is a 980ti with 12gb of vram. It's literally the same thing. Just has a different ID and no display out. Don't need benchmarks, it's the same thing.
@@raycert07 You wrote M40 instead of K40 (used in the video). Are you saying they're also the same? The price of the K40 is much less, so that would be interesting if they're the same. Thanks
@Christopher Robinson same series different generation. I was talking about the m40 because it's faster and it was tested in the video. The k40 is half as fast and uses a lot of power. It also has not been supported since 2021 and has clock speed issues. K40 is older. M40 is newer. K40 will have issues in modern games because it doesn't support dx12 and has no driver optimization. Don't buy Kepler. It's cheaper, but stop looking for cheaper, your not gonna benefit from the extra few dollars.
@Christopher Robinson last I checked the p40 is really expensive still, but m40 is really cheap. Maybe 70$ last I checked, but got mine a couple years ago for 150 which was still a steal.
I'm considering challenging that 8350 comment along with some of my future eWaste and what I have in mind. Recently I've become interested in gaming again. Maybe not to the point of cloud gaming but it still looks interesting. I want to take a low profile card like a P4, pair it to an FX-8370 and have that work as a dedicated 3D render and H264 stream encoding device. Those cards are now sub-$80 these days and they match the performance of my aging RX 580. It sounds like it can work.
Yes sir, thanks for the video, I like seeing these videos, lots of good older cards out there, Just go back to 1080P everyone until Crypt crashes again lol.
The card no longer sells for 100 bucks but I really hope those who got hands on one have a nice gaming experience now. It's just helpful for everyone as this hack massively reduces ewaste.
There are a few OEM Athlons, the 3150g seems to fit between the 3000g and the R3 3200g. Quad-core clocked similar to the 3200g but 192 shader cores like the 3000g.
Less than 3 weeks ago you could pick up the M40(12GB) for a little over $100(got mine for $120 each), now they're over $300. There are models of the k40 that do come with a blower fan, K40C, but those usually cost 3x as much. The "above 4g decoding" was just a lovely lesson to learn the hard way. My HP servers have a "Secret" menu option to enable it.(Ctrl+A in the bios) On a couple dell 'workstations', even with that option enabled(supposedly), it still won't boot. Really hit or miss on the older hardware.
I bought my m40 4 months ago and 90% of the listings were 150$. The k40c is NOT THE SAME AS THE K40. It's much much much worse. I actually think it's a Fermi card.
@@raycert07 They're back down to $120, and probably could best offer them lower. Side note, the 24GB variants are hovering around $200. I'll take your word on the k40c, don't personally own one, the roughly $150 price difference isn't worth it to just have a fan on it.
@@chrisbaker8533 the 24gb ones aren't worth buying anyways. No games can take advantage. It's been roughly 11 months since it's been watching prices and they have stayed stagnant until the last time I stopped looking around January. Most people look for the bundles, 200-300$ for a fan, printed shroud, power adapter, etc. Those are not worth it at all. I bought the gpu for 150, bought an adapter for 7$, and made my own cooling solution. At first I made a shroud out of taped paper with a blower fan, recently as of about 3 days ago u got tired of the noise so I took the cover off of the gpu, I previously removed the fins used for "cooling" the vrms and I made a new half shroud with a 80mm fan blowing down and into the heatsink. Much quieter and with a slight temperature increase. I would have hoped to have a cleaner solution like an aio on there, but I can't afford it right now. The vrm heatsink fins have been ripped off because they probably didn't do anything from what I could tell, and because it probably obstructed airflow to an extent.
@@chrisbaker8533 have you overclocked / bios modded your m40? If not I recommend you do. Temps aren't effected and performance makes it better than a 1070.
@@raycert07 Personally not a gamer, so my gpus are used for rendering, simulations and other technical stuff. I don't really overclock, as i need as high a level of stability as i can get. For cooling, i just ripped a couple fans out of an old r900 i had floating around and plugged them into the breakout board that powers the gpus. That was after i melted a couple pwm's, those fans pull some serious amps. I do agree that those shrouds and combo packs are way over priced. The fans they use are, top end, $10 bucks, and the shrouds are just 3d print. If you've already got a 3d printer, there's STL files all over the place and it might cost a couple bucks in material and electric. Noise isn't really an issue for me, as i have mine in a separate room. Just have to remind myself to put on hearing protection when i go in there. Don't really do any water cooling, just too many added failure points and risk for me to use it.
i scored a k80 with a bykski waterblock on it on ebay. im a total noob about watercooling tho. i need a radiator and some pipes and a pump, i guess.. wtf should i get?
Here in the UK, my GTX 980Ti that I bought second hand a year ago, just before the pandemic for £160 / $227 is now changing hands for more than double that.
ok i have a dell t7500 workstation with twin 5680s and have a k2200 video card plus a k80 . its so smashed full from the factory i tore everything out of the case and put it all into a cooler master haf 922 . now it has all kinds of room had to cut wires and slice some thing to get all to work but get it to go and is totally worth it ! to cool the k80 is very easy . you need to use 2 40mm base side out put water cooler blocks . i have a evga clc 120 water cooler ut one line and feed it to the first block . then connect the 2 with the out from the frist block to the in of the second block . then take the out from the 2nd block back to the hose you cut to complete the cooling . the evga fans sucked so i got a $20 Noctua NF-F12 PWM and it works awsome . it works so good that my temps never go over 90 degrees . so i put together the same thing for my cpus and now you cant even hear it run and cpus are at 80 degrees . to combat leaks i used brass barb fittings sealed with pipe dope and used house pex water pipe that i heated up to shrink to the fitting then clamped them . after that zero leaks system has been running for 2 years now non stop 24/7 !!!! gpu blocks $25 each www.amazon.com/gp/product/B07TP44HS6/ref=ox_sc_saved_image_3?smid=AET9B0FNKDKRC&psc=1 evga clc $40 www.amazon.com/gp/product/B07CNXHYJ2/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&th=1 pex www.homedepot.com/p/SharkBite-3-8-in-x-50-ft-Coil-White-PEX-B-Pipe-U855W50/202033028?g_store=&source=shoppingads&locale=en-US&&mtc=Shopping-BF-F_D26P-G-D26P-026_001_PIPE_FITTING-NA-NA-Feed-SMART-NA-NA-PIPE_AND_FITTINGS_New_Engen&cm_mmc=Shopping-BF-F_D26P-G-D26P-026_001_PIPE_FITTING-NA-NA-Feed-SMART-NA-NA-PIPE_AND_FITTINGS_New_Engen-71700000081569467-58700006932068304-92700062338492343&gclid=Cj0KCQiAgP6PBhDmARIsAPWMq6m5GBcOEqOYKtpT4JdlfxyZfAFE6iY4cG8BARJmajba-Nc0b3xWa6EaAmsEEALw_wcB&gclsrc=aw.ds last thing i forgot i use the program speedfan to control my cooling fan speeds
dayum, I was rocking the 908ti, very well I might add with a R7 2700x cpu. I got lucky with the 3000 series announcement and my coworker panic sold his 2080ti for 400$. Still have it on another pc and it's just a great card all around especially for 1080p, 2K on high you can still get high frames, last game I played on it was RE2. Just great.
wait, how does the Tesla K40 which is based on GK110, Keppler, aka basically 780ti, perform like the 980ti which is GM200, Maxwell which was like a 25% perf increase from Keppler to maxwell
The 780ti at 2gb was severely memory constrained vs 12gb on the tesla. Also this is overclocked if you follow the guide. "A" grade Kepler. It's close in real world game performance. The Maxwell version is even better.
@@Level1Techs thank you, would not have expected the high binning criteria would have been such a huge leap, id have expected between 780ti and titan performance at most things. really shows how far down the chain desktop consumers are normally. you got any numbers of what the K40 can get overclocked to on average roughly?
We only tried a few games so it's possible ymmv. It was a bigger breakthrough e.g. with Batman Arkham to get the physx driver working than the fps difference if that gives you any idea. Dirt was buggier on Kepler though
I miss your rants on the NSA and Governments... I stopped watching Tech syndicate when things went down... I have not been able to find another channel that is as formitive as you were back then... I feel the parting of ways has left a void for many people who want or is interested in that level of insight and information... Its to bad that you are not continuing with sharing your wisdom in that way... If you end up having a change of heart and bring that content in this channel, I forsure would be a loyal follower and listener to your words...
I was still rocking my 980TI playing at 1440p 144Hz until 2 months ago, sold it on ebay for £270, crazy price! But that is the way the second hand market is these days.
Am I missing something, but it seems that Tesla k40 doesnt have fan? I guess that K40 relies on server rack airflow for cooling? I would advise "gamers" to remove shroud from it and slap an 120mm zip-tied fan on it powered externally
Also make sure you set your pcie to 2.0 or 3.0 in your bios. If you have a board on 4.0 because you'll upgrade later and bought a newer mobo with that feature as a result, this will resolve *MANY* black or blue and frozen screens of death issues. Even a 3080 doesn't oversaturate a pcie 2.0x16 slot, so there will be little to no loss in framerates.
With reference to assigning physX, if one were to have multiple cards could you route the different processes needed to seperate GPUs then Funnell through to output¿
Have you ever played around with the Firepro S9050? Before the shortage hit, I was able to grab one new for $70 and was thinking of trying to pair it with an A-series APU (with the Dual graphics work around) or with the HD 7870 due to them being the same architecture. (Both are Tahiti, the 7870 is a 2gb while the S9050 is 12gb) Just wondering if there is FirePro version of your mad scientist ways.
This opened up so many options for gamers looking for a gpu right now. Also the amount you're essentially saving by buying a Tesla GPU or an ASIC would end up more than justify paying the extra premium for a cpu with an igpu or an apu. Well, at least until people start scalping these cards because now this information is in the wild and scammers gotta scam. lol.
Get those Tesla cards while you can, as soon as word of this gets widespread the price will probably at least double
Too late.
@@solocamo3654 i picked up 2 for 100 each but they are 17$ shipping so it better come bubbled wrapped so bad i can barely find it.
too late, the prices shot up tenfold....
@@tkpenalty Fuck...
Yep already up to $400... for the m40. The sold listings were about $100 before.
Its a pretty great card. With a modified inf file it can even be installed on XP. XP to windows 10 is pretty killer support.
NICE!
Oh hell yeah, time for a new budget build category
Yea
Apocalypse tier
Nah miners don't want them. Most only care about newer than Maxwell cards. Maxwell and older are prone to what's called dag thrashing or memory thrashing when attempting to mine a dagger hashimoto/ Ethereum algorithm. But those of us who want to play games or need a GPU to speed up whatever work we are doing, this is a great option.
"It's all going to heat". Wendell 6:7-8
The real question is how much MH's can it get? ))
@@sergeykoss not a great amount, for the power use. maxwell has severe deficits in architecture vs newer cards that make it pretty slow for mining even with 12 gigs of memory.
Maxwell used to be great for mining but in 2014-15. I got so many Dogecoin in its early days
I just love that Wendell tries to find ways to help out the gamers during the Great GPU Shortage of The 21st Century.
3:07 wow those are some spicy pillows.
S P I C Y -- replaced those in a video a while back LOL
I cannot resist to smash those pillows
Wait what laptop was in the video? Surface book? Surface laptop ?
Well RIP any decent prices. These are gonna go up now... Thanks a lot Wendell... 😂
Yeah, no kidding -- not $100 anymore. More like $250 or $300 now.
yeah that, but also poor students like me can start gaming in the graphics card crysis xdd
@@PoeLemic i see some just under 200 recently
Man, I'd love to see like Dawid or RandomGaminginHD or even Budget-Builds Official do performance reports/builds on these!
we are doing this
Same here. Would love to dawid does tech or BudgetGaminginHD
th-cam.com/video/sGd0Q9ocato/w-d-xo.html
@@awrsish that was just dumb. We might go live tomorrow on miyconst channel
NO. Last time this happened was with the p106 cards and what happened when someone pointed out you can game on them? Nvidia blocked the feature and with the lack of driver support for games, you can't play anything. We need to keep this on the down low so the people that know can enjoy it. And if larger channels cover it, prices will skyrocket from 150$ to 400$, it's not even worth it. It needs to be an inside secret. Once nvidia catches on because more people report on it, all support is gone, graphics mode is gone, drivers are gone, no gaming. A expensive blender card.
You don't need performance reports or builds. Build your own, do your own testing. It's a titan x. It performs the same as a 1070. That's all you need to know for game performance. Don't bring too much attention to it before you ruin it.
i managed to get Tesla K20Xm to work with GTX Titan bios, enabling full OC support and making it work in Windows 10 without any modifications to the OS itself.
Matter of modifying device id straps in Titan bios so that it gets detected as a Quadro GPU, thus enabling raster rendering by default. And I used Morpheus II cooler on it, fits nicely.
that is great
Awesome stuff Wendell.
Glad you mentioned the X300 Deskmini again, I picked one of those up shortly after your video on it and paired it with a 4650g and I have to say it has been a wonderful Home Theater/Living room PC.
On cooling these, the Kraken G12 with a compatible AIO will handle that without a problem, especially if you use a dual-120mm AIO (e.g. the Kraken X53). I have two Tesla M40s that I swap back and forth between Folding@Home and some BOINC projects. The AIOs keep those processors cool without any issues. The memory will run hot, but directed airflow can help with that.
At minimum, redo the thermal paste on the GPU as it's likely completely hardened over.
This card is nice! The price is gonna go up though.
Already did.
Wendell is to computer hardware what Elon is to bitcoin.
@@Arnau975 any tech person on YT with a viewership in thousands can do that. The dude from Craft Computing doubled the prices of the Sun F80 SSDs on ebay the same way.
I was gonna ask where you find this card for $100 lol. Maybe last year .
@@ivchatov I'm still sour about that one...
"We are going to hack windows"
I was sold right there xD
I ended up getting 4 K80s to pair with an x99 board with a 5960X CPU on it, after tweaking a lot of BIOS settings. Asus x99 Deluxe II I think? It does run them at pcie2 for bandwidth since the K80 are dual GPU cards, but I'm hoping to get cooling working on it all soon and have a decent rendering/computational box for cheap
how much did you pay for them ?
With servers there are usually really loud fans. Just ask anyone who has ever powered one on and they might tell you they are NOT silent. It is for the simple fact that the components of a server don't have their own fans for a number of reasons.
Nice video man! I'm glad to see a tech TH-camr showing options instead of just showing how great the new cards are that you can't even find to buy anywhere.
I have been suggesting to people to get the K80 which can be had for 200-300 depending on where you live. It is close enough to a modern GPU that you can do decent gaming. You should also be able to use multiple k80s in a system.
Why would you do that. The k80 is just 2 k40s on one card. You can't use sli properly either and you can't use all 24 gb because with sli, vram does not stack. Even with sli, it only performs the same if not worse than a m40, a gpu that you can get for 150$. The k80 performs like 2 1060s in sli which only sometimes matches the m40 in performance. Not to mention kepler has lost driver support, uses more power, has no reflex support, and is generally worse. It's not a good deal even for 200$
Also, multiple k80s? The k80 is a dual gpu card, you can't sli a gpu already internally in sli.
G'day from Sydney. Cool idea. Market forces in Australia - population 25 million - unfortunately make this cost prohibitive (had a glance at secondary market, yikes) to try as a lark.
Was looking at these Tesla gpus for octane rendering a few weeks ago as the cuda count is still pretty good and that 12gb would be nice for heavy scenes.Had read they used to be and are still used in a few render farms. A few bios issues meant I couldn't / wouldn't be able to get them to work which you mentioned. Thanks for noting and testing that.
thanks for the z77 disclaimer, i was about to pull the trigger on this for my old 3770k!
On your system you can likely get full Windows XP support along with getting a 970, 980ti, and TitanX running by modifying the 960 driver's ini.
I remember that I was kicked out from a tech forum many years back for claiming that you could in theory could game on a Tesla if youre able to hack the software.
Guess I was right after all
This reminds me of Lucid's Virtu. They used to use an abstraction layer to intercept and redirect api calls to either discreet or inbuilt GPU. Though it's target was to get quick sync to work even when discrete graphics is attached, as intel used to disable igpu when a discrete card was attached. Funny part is when benchmarking . Dirt series (Dirt 2 back then) performed really poorly on that too.
I bought a 980ti kingpin from evga bstock last summer. 99$ and is a world record bin card. Last I checked I was #2 on stock cooling with it on 3dmark
I've been using this with a dual gpu card where the first GPU (which has the display outputs) doesn't work anymore. Now it outputs the frames from the second gpu through the Intel HD graphics.
Were you trying to make it hopelessly inefficient or just ended up that way?
@@unitedfools3493 No, I bought this Radeon Pro Duo as defective, but I managed to still use it for something with this workaround. That something being a gaming PC for the living room TV.
Good on ya for keeping it at least somewhat functional instead of letting it become waste! Any use you can get out of it is better than it going to the trash.
Actual eBay price: 295 to 800$. Srsly wtf is going on.
3 weeks ago you could find them for 90-120. but then everyone caught on, and videos like this made those deals go extinct
I picked up a k40m for 40 bucks im waiting on it to come i will update once i get it setup ill will edit it and tell you all how it went
Wendel Every Video you make is so relaxing and so informative. Thank you for all your knowledge Sir!!
Thank you for turning a thought I had into an actual video, now I know its a real thing.
Same :D
My GTX 960 although only having 2 GB of memory on it was not expensive and it still runs everything I ever wanted. I am not one of those guys looking for insanely high frame rates but more or less just playing my games at good frame rates.
Last December, i bought a tesla P4 which cost for 60 usd. Regular price was around 92 USD. I got it cheap because the online shop that i've is celebrating a Christmas to new year's eve sale. I am really happy with it. Paired it up with an Lenovo SFF i7-7700 and can play any games at 1080p with low and some with meduim. I can even play God of war III in RPCS3 at 720p. It dips down below 60 fps, even 30 fps but really playable
Time for me to buy two and try and sli them....
Wendell was buying these up weeks before releasing this ;)
Sold, thanks Wendel! I'm assuming that it would work the same way with looking glass?
I believe you would have to passthrough 2 GPUs, the Tesla and the igpus, I could be wrong, not Linux strong
Ive got an K20m and will defiantly try it out and post it on the forums.
Nvidia need to do an ampere refresh of the tesla ASAP. Would help a lot with the mining and scalping problems we have at the moment.
Could you do a follow Up Video for Tesla M40 Cards? In the forums there is evidence of them working the same way as you describe for the K80s and since M40 basically is a GTX TITAN X (Maxwell 2.0), I am sure many are interested in using these cards, too... Great video, btw :)
I just put together a Frankenstein contraption a month ago. It's a 2017 Alienware Aurora R6 with an intel i7 7700k, a GTX 1080fe and a Quadro K6000. The hurdle was uninstalling all graphics software and installing graphics software from 2017-2018. Once I did that, Windows 10 recognized both cards at the same time. Better still,..Blender recognised and is running on the K6000 card with 12GB Vram rather than the 1080 card with 8GB Vram. How or why that is happening I'll never figure out because from what I have researched,..Blender utilizes the lowest card available rather than the highest,..so??? Now I'm curious what to expect when I rebuild and use the K6000 with the Tesla K80 I found on E-Bay. P.S. I have less than $1,000.00 invested in the K6000 and the K80. I can't buy a RTX 3060 for that. I also upgraded my power supply to a Corsair RM850X 80 Gold and installed 64GB Corsair Vengeance DDR5 RAM. I'm not a gamer so I can't tell anything about gaming performance. I can tell my rendering time in Cycles decreased by a factor of at least three and according to thermals nothing is even getting warm under a heavy load render.
so i have tried the guid and tried flashing a GTX TITAN vbios to my tesla K20Xm card and nothing works... apparently all graphics cards are not equal. can you possibly do a video or series on how to make the tesla card a titan?
Nice video. Thou you cannot find tesla k40 anywhere in "europe" for atleast less than 200e
have u tryed using a grid card instead of the telsta seen grid cards sit in that 150 range seem like they could be a little better game features wise out of box
The other way around would be interesting. I would like to use my 980ti (when I replaced it with a new gpu in 2025 🙄 ) for my tf object detection .
Yeah, that's when the new GPU's are gonna be affordable. I'm just waiting -- they are too much now for the common man.
Iris Xe on laptops was a bit more impressive than I thought, I mean the same laptop has an RTX 3070 (Secretly Max-Q). The Xe Graphics, when playing the same game, ran at 100% ad gave a very playable performance of 50FPS, on high settings (1080p with SR 1.2) . Lowering the settings probablies would give a more competitive frame rate, and the D-GPU is more power efficient and would give a better performance, even with RTX on medium.
If you can find that 3000g for msrp. I paid 80 bucks for a pc I was building for my in laws. Smh. The market is totally messed up.
Aight I may upgrade later this year. To a Ryzen APU with a Tesla card. Currently have a Ryzen 3600 and I'm waiting on a 980 in the mail.
this might be one of the few channels where releasing this info won't instant sell out and go back for 2x the price on ebay.
What would be better for gaming, A K40 or the K80?
Love the new intro animation
Thank you :) our other editor made it! ~ Editor Amber
L1 just saving the planet much credit deserved for flexing them brain muscles and working to game and be productive
Where my 980ti people? We got a good excuse to keep it even longer lol
I still love my 980ti FTW. Still a beast for 1080 gaming.
@@carth85 about to water cool mine! Just for temps, seeing 70-85+ even with 100% fan !!
just bought a k20x for 125 CAD plus hipping and tax im at 180 CAD for a GTX titan that Cost 10K plus in its day. im a happy boy .Gonna try all my trick to get it working . Its headless but i can use anything
I can confirm this working on z97 asus mobo and a i5 4590 and a TESLA k40m, followed the instructions and installation, and overclocking worked, MSI AFTERBURNER IS NEEDED. It is noted in the instructions, that you need to run the nvidia-smi commands after every reboot, however if you shutdown you won't need to reenter the commands, with rebooting yes you do. Edit: forgot to add the GPU temps with 3d printed fan included Blower style cooler bought from ebay for about 35 dollars, less than 60c running heaven benchmark, at 1125 mhz core clock, and 3125 mhz memory clock, WITHOUT changing thermal paste or cleaning the card.
edit2: overclocked the Tesla k40 to 3154MHZ Memory, and 1200MHZ Core Clock ran Heaven benchmark twice no errors, and temps were under 60c the whole time I am going to clean and replace the thermal paste and any thermal pads on the cards to see if they drop further, NOW the caveat is the blower fan is LOUD but the gaming performance is SOLID!!!
Any updates? Very similar to what my plans are. Still working nicely enough?
after reading the comments, apparently the M40 was available as low as ~$100 a couple weeks ago, but now its up to ~$300+. M40 would make sense for 980 Ti level performance, but shouldn't a K40 for $100 be more like 780 Ti performance? And of course it would have the pitfalls of Kepler being a bit too long in the tooth, where newer games like Doom Eternal can run like garbage regardless of the decent raw power of the GPU
The nvidia smi and clocks make it benchmark about the same as a 980/Ti. Don't forget you have 12g vram and the 780 2gb wasn't great. This is higher clocked.
It'd be great if you could post some benchmark numbers to see how they compare to a 980ti, I'm really thinking about trying it only because I like to experiment with stuff.
Great video!
It's literally a 980ti. The m40. It's the same dye, it's actually closer to a titan x, which is a 980ti with 12gb of vram. It's literally the same thing. Just has a different ID and no display out. Don't need benchmarks, it's the same thing.
@@raycert07 You wrote M40 instead of K40 (used in the video). Are you saying they're also the same? The price of the K40 is much less, so that would be interesting if they're the same. Thanks
@Christopher Robinson same series different generation. I was talking about the m40 because it's faster and it was tested in the video. The k40 is half as fast and uses a lot of power. It also has not been supported since 2021 and has clock speed issues.
K40 is older. M40 is newer. K40 will have issues in modern games because it doesn't support dx12 and has no driver optimization. Don't buy Kepler.
It's cheaper, but stop looking for cheaper, your not gonna benefit from the extra few dollars.
@@raycert07 thanks, will look into M40 and p40
@Christopher Robinson last I checked the p40 is really expensive still, but m40 is really cheap. Maybe 70$ last I checked, but got mine a couple years ago for 150 which was still a steal.
I'm considering challenging that 8350 comment along with some of my future eWaste and what I have in mind. Recently I've become interested in gaming again. Maybe not to the point of cloud gaming but it still looks interesting. I want to take a low profile card like a P4, pair it to an FX-8370 and have that work as a dedicated 3D render and H264 stream encoding device. Those cards are now sub-$80 these days and they match the performance of my aging RX 580. It sounds like it can work.
Yes sir, thanks for the video, I like seeing these videos, lots of good older cards out there, Just go back to 1080P everyone until Crypt crashes again lol.
Well, I can't wait until Cryptos crash. I sure think they will.
The card no longer sells for 100 bucks but I really hope those who got hands on one have a nice gaming experience now.
It's just helpful for everyone as this hack massively reduces ewaste.
The k40 is closer to a 1060, nearly matching a 1060 actually, and surpassing it when overclocked. But the m40 is actually a 980ti / titan x.
You're awesome! I was wondering the answer to this exact question al throughout the holidays
i get excited when i see the 3000G in stock at Microcenter... hopefully some day there will be a Zen 2/3 version of the 3000G, too!
*SOME DAY *SOON*
There are a few OEM Athlons, the 3150g seems to fit between the 3000g and the R3 3200g.
Quad-core clocked similar to the 3200g but 192 shader cores like the 3000g.
@@thepezfeo wait whaaaaat? where is this?? Aliex only or something?
Less than 3 weeks ago you could pick up the M40(12GB) for a little over $100(got mine for $120 each), now they're over $300.
There are models of the k40 that do come with a blower fan, K40C, but those usually cost 3x as much.
The "above 4g decoding" was just a lovely lesson to learn the hard way.
My HP servers have a "Secret" menu option to enable it.(Ctrl+A in the bios)
On a couple dell 'workstations', even with that option enabled(supposedly), it still won't boot.
Really hit or miss on the older hardware.
I bought my m40 4 months ago and 90% of the listings were 150$. The k40c is NOT THE SAME AS THE K40. It's much much much worse. I actually think it's a Fermi card.
@@raycert07 They're back down to $120, and probably could best offer them lower.
Side note, the 24GB variants are hovering around $200.
I'll take your word on the k40c, don't personally own one, the roughly $150 price difference isn't worth it to just have a fan on it.
@@chrisbaker8533 the 24gb ones aren't worth buying anyways. No games can take advantage. It's been roughly 11 months since it's been watching prices and they have stayed stagnant until the last time I stopped looking around January.
Most people look for the bundles, 200-300$ for a fan, printed shroud, power adapter, etc. Those are not worth it at all. I bought the gpu for 150, bought an adapter for 7$, and made my own cooling solution. At first I made a shroud out of taped paper with a blower fan, recently as of about 3 days ago u got tired of the noise so I took the cover off of the gpu, I previously removed the fins used for "cooling" the vrms and I made a new half shroud with a 80mm fan blowing down and into the heatsink. Much quieter and with a slight temperature increase. I would have hoped to have a cleaner solution like an aio on there, but I can't afford it right now. The vrm heatsink fins have been ripped off because they probably didn't do anything from what I could tell, and because it probably obstructed airflow to an extent.
@@chrisbaker8533 have you overclocked / bios modded your m40? If not I recommend you do. Temps aren't effected and performance makes it better than a 1070.
@@raycert07 Personally not a gamer, so my gpus are used for rendering, simulations and other technical stuff.
I don't really overclock, as i need as high a level of stability as i can get.
For cooling, i just ripped a couple fans out of an old r900 i had floating around and plugged them into the breakout board that powers the gpus.
That was after i melted a couple pwm's, those fans pull some serious amps.
I do agree that those shrouds and combo packs are way over priced.
The fans they use are, top end, $10 bucks,
and the shrouds are just 3d print.
If you've already got a 3d printer, there's STL files all over the place and it might cost a couple bucks in material and electric.
Noise isn't really an issue for me, as i have mine in a separate room.
Just have to remind myself to put on hearing protection when i go in there.
Don't really do any water cooling, just too many added failure points and risk for me to use it.
i scored a k80 with a bykski waterblock on it on ebay. im a total noob about watercooling tho. i need a radiator and some pipes and a pump, i guess.. wtf should i get?
Here in the UK, my GTX 980Ti that I bought second hand a year ago, just before the pandemic for £160 / $227 is now changing hands for more than double that.
Tensorflow and pytorch benchmarks on Kepler vs pascal!!!!
Raises one eyebrow; "Fascinating." Can't wait to try this!
Well, don't know if its cause of this video but the K40 is currently priced at above 220$+ and even the K10 is at 120$+ on eBay right now.
im still using the 980 TI and it works! it works pretty good!
ok i have a dell t7500 workstation with twin 5680s and have a k2200 video card plus a k80 . its so smashed full from the factory i tore everything out of the case and put it all into a cooler master haf 922 . now it has all kinds of room had to cut wires and slice some thing to get all to work but get it to go and is totally worth it ! to cool the k80 is very easy . you need to use 2 40mm base side out put water cooler blocks . i have a evga clc 120 water cooler ut one line and feed it to the first block . then connect the 2 with the out from the frist block to the in of the second block . then take the out from the 2nd block back to the hose you cut to complete the cooling . the evga fans sucked so i got a $20 Noctua NF-F12 PWM and it works awsome . it works so good that my temps never go over 90 degrees . so i put together the same thing for my cpus and now you cant even hear it run and cpus are at 80 degrees . to combat leaks i used brass barb fittings sealed with pipe dope and used house pex water pipe that i heated up to shrink to the fitting then clamped them . after that zero leaks system has been running for 2 years now non stop 24/7 !!!!
gpu blocks $25 each
www.amazon.com/gp/product/B07TP44HS6/ref=ox_sc_saved_image_3?smid=AET9B0FNKDKRC&psc=1
evga clc $40
www.amazon.com/gp/product/B07CNXHYJ2/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&th=1
pex
www.homedepot.com/p/SharkBite-3-8-in-x-50-ft-Coil-White-PEX-B-Pipe-U855W50/202033028?g_store=&source=shoppingads&locale=en-US&&mtc=Shopping-BF-F_D26P-G-D26P-026_001_PIPE_FITTING-NA-NA-Feed-SMART-NA-NA-PIPE_AND_FITTINGS_New_Engen&cm_mmc=Shopping-BF-F_D26P-G-D26P-026_001_PIPE_FITTING-NA-NA-Feed-SMART-NA-NA-PIPE_AND_FITTINGS_New_Engen-71700000081569467-58700006932068304-92700062338492343&gclid=Cj0KCQiAgP6PBhDmARIsAPWMq6m5GBcOEqOYKtpT4JdlfxyZfAFE6iY4cG8BARJmajba-Nc0b3xWa6EaAmsEEALw_wcB&gclsrc=aw.ds
last thing i forgot i use the program speedfan to control my cooling fan speeds
They should sell GPUS in external enclosures, maybe that will help them design around supply issues.
dayum, I was rocking the 908ti, very well I might add with a R7 2700x cpu. I got lucky with the 3000 series announcement and my coworker panic sold his 2080ti for 400$. Still have it on another pc and it's just a great card all around especially for 1080p, 2K on high you can still get high frames, last game I played on it was RE2. Just great.
the question how will these graphics cards work in a egpu dock IE the razer core?
Please someone can help me , does this procedure would work on a Tesla k20 5gb (the version with active cooling ) ???
Question, isnt the gpu in the 2400/3400g better than the 4750G, the 4750G is a better cpu as far as i know
I've just started enjoying games that run on potatoes. I have both Cyberpunk 2077 and Factorio. Factorio is a lot more fun.
3:06 yeah those batteries look a little bloated.
This GPU can be virtualized, literally worth the price.
Sheeesh,u here 2?
Like, you add it to a virtual machine in VMware or u use a Linux vm
Only if you do some diy stuff or pay nvidia 4k$ per vm per gpu per year.
@@EugeneBuvard no. VMware sucks. And linux vm does not make sense. You have to use gpu pv or pcie pass-through on linux.
@The Stock Trader explain what. Gpu passthrough is extremely complicated
I have Tesla K10 running as CUDA addition to my HP Zbook 17 under eGPU TB3. I might try this and seek for more mods so it will perform better.
wait, how does the Tesla K40 which is based on GK110, Keppler, aka basically 780ti, perform like the 980ti which is GM200, Maxwell which was like a 25% perf increase from Keppler to maxwell
The 780ti at 2gb was severely memory constrained vs 12gb on the tesla. Also this is overclocked if you follow the guide. "A" grade Kepler. It's close in real world game performance. The Maxwell version is even better.
@@Level1Techs thank you, would not have expected the high binning criteria would have been such a huge leap, id have expected between 780ti and titan performance at most things. really shows how far down the chain desktop consumers are normally. you got any numbers of what the K40 can get overclocked to on average roughly?
We only tried a few games so it's possible ymmv. It was a bigger breakthrough e.g. with Batman Arkham to get the physx driver working than the fps difference if that gives you any idea. Dirt was buggier on Kepler though
Ive been gaming with an intel hd 3000 integrated for a while now, its a pain in the ass but I can get by with it.
Checked Ebay......... Some starting at $150...... the rest are ridiculous prices.
I miss your rants on the NSA and Governments... I stopped watching Tech syndicate when things went down... I have not been able to find another channel that is as formitive as you were back then... I feel the parting of ways has left a void for many people who want or is interested in that level of insight and information... Its to bad that you are not continuing with sharing your wisdom in that way... If you end up having a change of heart and bring that content in this channel, I forsure would be a loyal follower and listener to your words...
Those cost like 200-300€ in Germany since 2-3 Years and prices never dropped here towards the 100€ mark 😅
i HAD A 980 unitl a15 months ago and it worked well.
A ti SHOULD BE FINE TODAY.
How well does it work with Looking Glass and Nvidia FBC codec? Iirc this feature is (officially) restricted to professional GPUs.
Are you kidding me??? I'm still rocking a 980ti and it's still holding fine on my haswell 4690 (non k) this is not a joke. My system still runs great
I was still rocking my 980TI playing at 1440p 144Hz until 2 months ago, sold it on ebay for £270, crazy price! But that is the way the second hand market is these days.
Actually, Quadro Pascal and more recent Quadros have such massive jump from previous generation. I'm even using Quadro P620 for my HTPC / Slim PC.
Am I missing something, but it seems that Tesla k40 doesnt have fan? I guess that K40 relies on server rack airflow for cooling? I would advise "gamers" to remove shroud from it and slap an 120mm zip-tied fan on it powered externally
Hence the entire section of the video and forum post talking about 3d printed fan adapters
@@Level1Techs yeah. For those who dont have 3D printer, one could tinker with improvised means of cooling. Dremel's up! =)
Good reminder. Just because it doesn't have a display out doesn't mean it isn't a GPU. If you can path an output give it a whirl.
I'm wondering about using it for PCI-passthrough on a Linux host, using looking glass? Or, if it isn't, would it be useful for VirGL?
Also make sure you set your pcie to 2.0 or 3.0 in your bios.
If you have a board on 4.0 because you'll upgrade later and bought a newer mobo with that feature as a result, this will resolve *MANY* black or blue and frozen screens of death issues.
Even a 3080 doesn't oversaturate a pcie 2.0x16 slot, so there will be little to no loss in framerates.
With reference to assigning physX, if one were to have multiple cards could you route the different processes needed to seperate GPUs then Funnell through to output¿
Tesla K40 is about $200 on ebay right now. Here in Russia though, they're still sold in stores for like $1600 lmao.
Me still rocking my 980 ti because of the shortage >.>
Same here 🔥 still got my 3770K too 😒😂
We don't have the option of cheap Tesla cards here, all options in online market places are overpriced.
i bought my k40 a while ago by literally just bidding $50 on a bid, forgetting about it, and surprising myself by it arriving in the mail
Have you ever played around with the Firepro S9050? Before the shortage hit, I was able to grab one new for $70 and was thinking of trying to pair it with an A-series APU (with the Dual graphics work around) or with the HD 7870 due to them being the same architecture. (Both are Tahiti, the 7870 is a 2gb while the S9050 is 12gb) Just wondering if there is FirePro version of your mad scientist ways.
This man knows this stuff. Love you videos :)
Stupid question. Why aren't these cards being bought by miners?
Prices have already skyrocketed. No longer worth it, as a the consumer versions are only marginally more expensive. :(
look up all the keplar cards possible for this. a k2, for instance, might work
This opened up so many options for gamers looking for a gpu right now.
Also the amount you're essentially saving by buying a Tesla GPU or an ASIC would end up more than justify paying the extra premium for a cpu with an igpu or an apu. Well, at least until people start scalping these cards because now this information is in the wild and scammers gotta scam. lol.
I really don't see a Kepler gpu pushing 980 ti performance. Did you mean 780 ti maybe?