Gaming On An Old Tesla GPU: 980ti Performance for $100?!
Vložit
- čas přidán 12. 05. 2021
- forum.level1techs.com/t/gamin...
**********************************
Thanks for watching our videos! If you want more, check us out online at the following places:
+ Website: level1techs.com/
+ Forums: forum.level1techs.com/
+ Store: store.level1techs.com/
+ Patreon: / level1
+ KoFi: ko-fi.com/level1techs
+ L1 Twitter: / level1techs
+ L1 Facebook: / level1techs
+ L1/PGP Streaming: / teampgp
+ Wendell Twitter: / tekwendell
+ Ryan Twitter: / pgpryan
+ Krista Twitter: / kreestuh
+ Business Inquiries/Brand Integrations: Queries@level1techs.com
IMPORTANT Any email lacking “level1techs.com” should be ignored and immediately reported to Queries@level1techs.com.
-------------------------------------------------------------------------------------------------------------
Intro and Outro Music By: Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
creativecommons.org/licenses/b... - Věda a technologie
Get those Tesla cards while you can, as soon as word of this gets widespread the price will probably at least double
Too late.
@@solocamo3654 i picked up 2 for 100 each but they are 17$ 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.
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 ?
I just love that Wendell tries to find ways to help out the gamers during the Great GPU Shortage of The 21st Century.
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 $300 now.
yeah that, but also poor students like me can start gaming in the graphics card crysis xdd
@@PoeLemic i see some just under 200 recently
"It's all going to heat". Wendell 6:7-8
The real question is how much MH's can it get? ))
@@sergeykoss not a great amount, for the power use. maxwell has severe deficits in architecture vs newer cards that make it pretty slow for mining even with 12 gigs of memory.
Maxwell used to be great for mining but in 2014-15. I got so many Dogecoin in its early days
"We are going to hack windows"
I was sold right there xD
Man, I'd love to see like Dawid or RandomGaminginHD or even Budget-Builds Official do performance reports/builds on these!
we are doing this
Same here. Would love to dawid does tech or BudgetGaminginHD
czcams.com/video/sGd0Q9ocato/video.html
@@awrsish that was just dumb. We might go live tomorrow on miyconst channel
NO. Last time this happened was with the p106 cards and what happened when someone pointed out you can game on them? Nvidia blocked the feature and with the lack of driver support for games, you can't play anything. We need to keep this on the down low so the people that know can enjoy it. And if larger channels cover it, prices will skyrocket from 150$ 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.
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.
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
Wendel Every Video you make is so relaxing and so informative. Thank you for all your knowledge Sir!!
You're awesome! I was wondering the answer to this exact question al throughout the holidays
Nice video man! I'm glad to see a tech CZcamsr showing options instead of just showing how great the new cards are that you can't even find to buy anywhere.
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 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 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¿
Wow this was fantastic! Loved this technical hack, super cool. Subscribed.
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.
@@raycert07 Can someone explain this to me?
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.
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, $10 bucks,
and the shrouds are just 3d print.
If you've already got a 3d printer, there's STL files all over the place and it might cost a couple bucks in material and electric.
Noise isn't really an issue for me, as i have mine in a separate room.
Just have to remind myself to put on hearing protection when i go in there.
Don't really do any water cooling, just too many added failure points and risk for me to use it.
On cooling these, the Kraken G12 with a compatible AIO will handle that without a problem, especially if you use a dual-120mm AIO (e.g. the Kraken X53). I have two Tesla M40s that I swap back and forth between Folding@Home and some BOINC projects. The AIOs keep those processors cool without any issues. The memory will run hot, but directed airflow can help with that.
At minimum, redo the thermal paste on the GPU as it's likely completely hardened over.
I'm wondering about using it for PCI-passthrough on a Linux host, using looking glass? Or, if it isn't, would it be useful for VirGL?
Actual eBay price: 295 to 800$. 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
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.
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...
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.
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.
This man knows this stuff. Love you videos :)
Have you ever played around with the Firepro S9050? Before the shortage hit, I was able to grab one new for $70 and was thinking of trying to pair it with an A-series APU (with the Dual graphics work around) or with the HD 7870 due to them being the same architecture. (Both are Tahiti, the 7870 is a 2gb while the S9050 is 12gb) Just wondering if there is FirePro version of your mad scientist ways.
I have been suggesting to people to get the K80 which can be had for 200-300 depending on where you live. It is close enough to a modern GPU that you can do decent gaming. You should also be able to use multiple k80s in a system.
Why would you do that. The k80 is just 2 k40s on one card. You can't use sli properly either and you can't use all 24 gb because with sli, vram does not stack. Even with sli, it only performs the same if not worse than a m40, a gpu that you can get for 150$. 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.
Do the newer Tesla cards still have all the graphics fixed function blobs? Or did they use the space for more CUDA and Tensor units, like on CDNA?
Is it possible to use one of these cards in a windows vm and get the output through looking-glass?
that got me interested, is this something viable for someone doing renders? my specific config is a r5 1600 in a B450 chipset currently using a 1050ti as display output and maybe wondering to use an old tesla with lots of vram to speedup my blender renders
Thank you for turning a thought I had into an actual video, now I know its a real thing.
Same :D
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
thanks for the info .. would you please explain the drivers setup once more ? i have a tesla k80 and a quadro k6000 .. ill have to install the tesla alone 1st and intall its driver , then install the quadro and its driver after ?
Not surprised... But very happy of the in depth review and workaround kind of video. Kind of makes you think if you buy a quaddro that you buy a low powered gaming card with a bit better support and some "not always" better drivers
Ive got an K20m and will defiantly try it out and post it on the forums.
Stupid question: Could this be added as a support card for a dedicated GPU in normal desktop pc circumstances, ilke you got you classic GPU in the first PCI slot and then put one of these in the second slot and get some more performance?
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.
Do you think we could use this to get SR-IOV "working" with GeForce cards?
Use a cheap Quadro with grid and pass the frame buffer of a GeForce card to the Quadro which in turn gets used by the VMs.
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.
you ever figure out a way to flash bios or something to overclock the m40? (ive found waterblocks that will fit this)
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
Wendell was buying these up weeks before releasing this ;)
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
Do you have to have a integrated igp to transer the graphics too? I have a 1050ti and can i use that as the gpu to transer too?
the question how will these graphics cards work in a egpu dock IE the razer core?
Hello , great channel , I always visit here for technical info . Regarding this Tesla card I have one with 12gb , can I use this one with Quadro k6000 or m4000 on dell or HP workstation with dual Cpu ?
My work is mainly related with 3D computation and video . Since this card does not have display . Little confused how to go about it .
You should be able to
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 !!
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 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?
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.
How can use this Tesla on a server direct win install or with a VM with windows 11 for GIS / CAD and get monitor output..via ethernet??
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?
Love the new intro animation
Thank you :) our other editor made it! ~ Editor Amber
Thank you. Really... thank you ❤❤
does this work when running in a VM with passthrough, using the virtual GPU as the weak one? I guess at that point though, you're just doing VDI with extra steps.
No. The virtual gpu is emulated. It will render and pass it "through" but it will be laggy looking and won't be good at all because the output is emulated.
Please someone can help me , does this procedure would work on a Tesla k20 5gb (the version with active cooling ) ???
Hello. I have a LEADEX III GOLD 850W power supply. I bought a Quadro M6000 graphics card. The video card consumes up to 250W. There is only one 8Pin connector for additional power supply on the video card. Can I power the video card with one 8Pin PCIe connector from the power supply? One plug on the power supply is designed for such a current? Or do I need a "PCIe Dual 6-Pin to 8-Pin Adapter Cable ( two 6-pin auxiliary PCIe connectors into a single 8-pin auxiliary
PCIe connector))"? And I have to power from two PCIe connectors on the PSU through this adapter?
Wendell does this work on grid gpus as well?
Hi, is it possible to mount a Tesla GPU to a thunderbolt 3/4 platform for eGPU application to accelerate RTX3060 and boost VRAM? on a Acer Nitro 5?
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 :)
Any idea of a GT 1030 instead of an iGPU? Most X299 boards don’t have display outputs, and most extreme edition chips have iGPUs switched off. The ROM BAR and 4G decode shouldn’t be an issue. I have a Tesla K80 I want to run on an X299 board with an i7-7820X.
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?
Can I use this with my gaming video card in sli only for the VRAM?
I have a 1080 cooler I wonder if it would work on this card. May have to buy one and give it a go
Any help would be greatly appreciated. I have a hyper-v 2016 server running on a Dell R740xd server, and I need 6 Windows vm's (connected to using RDP from WYSE 3040 thin client) to be able to play CZcams training videos with decent quality, nothing fancy and no need for any gaming specs. Microsoft removed RemoteFX so I can't use that anymore and now they recommend DDA, whatever that is. I am not sure if the free VMware ESXi server will work either. Any solution you may have would be greatly appreciated.
Wauw thats very cool. Pitty i didn't try this before i true
them, once a verry expensive company card, a way :-( I also wonder what you can still do with Quardro cards like die P4000 series...never mind true them i way to :-)
Tensorflow and pytorch benchmarks on Kepler vs pascal!!!!
Thank you
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.
been trying something similar with a grid k2 i picked up for 94 usd it kinda works when svga is present but the game im trying to run doesnt like anything vmware so idk i would have to learn something else for the specific game but all games without that check should work under esxi
where is the list of compatible mobos?
you mentioned X99 II but not X99 I? does that include X99-A II
Is there any guide on how to create an automate script after reboot so I won't do the cmd Nvidia smi every time?
Could the bios be replaced with a 980ti bios if you had the tools ?
Great video!!! I've been thinking about what to do with 2 Teslas c1060 I got out of eBay 10 years ago..... Any ideas? Cooling is already figured out..... Crypto Mining anyone? Any other computational intensive activities? Any software out that can make good use of them?
so the problem is the amount of vram if i use for example an tesla k20x with 6gb of vram i can use with out problem on a chinese x79 motherboard?
If I assign such a card to a KVM-VM would it be possible to run Gaming that way?
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.
What would be better for gaming, A K40 or the K80?
i'm sorry, i'm a bit dumber. at 5:00 you say that if you add a quadro to your pc, your gtx will act the same way as the quadro.
how does the quadro act and what changes on the geforce?
Time for me to buy two and try and sli them....
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.
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.
Can these cards be used with PCI-E passthru + KVM + looking glass?
yes.
you are gonna need to do some setting up though.
How well does it work with Looking Glass and Nvidia FBC codec? Iirc this feature is (officially) restricted to professional GPUs.
question: should i buy a tesla p100 or sli the tesla m40
Neat. Not something I have a need for but probably wouldn't work in my system anyway. I've tried enabling above 4G decoding on my X370 board before to see whether it makes any difference (doubt it) and it causes both the onboard LAN controller and NVMe drive in the second M.2 slot to not get recognised by Windows (and likely more issues I didn't immediately notice). No idea how you'd got about fixing that. Didn't try a fresh Windows installation on a board that had it enabled already, maybe that would work. Also means that even if SAM/rBAR was supported by mainboard and CPU it wouldn't actually work when above 4G decoding breaks stuff.
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.
Raises one eyebrow; "Fascinating." Can't wait to try this!
What about the Pascal based P106 mining cards ? Most people that have tried, have failed in 1 way or another.
Would that work well for video editing or design? As a secondary gpu
Design is too vague and video editing it probably works
nvenc encoding for obs? is it compatible?
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.
Well, you could use the nvme connector to use a pci-e riser and external gpu on thay desk mini. Maybe look into that?
I've heard that it doesn't play well with risers, but those people didn't know much about the gpu anyways so it just might work. I want to try this on a laptop but I lack the funds to do so right now.
I got my hands on a Quadro P4000. Do you think there's a similar method to this to allow it to behave similarly?
Yes. Exactly the same steps in theory but these steps are flawed. There is a reddit post that you can find that does it way better. Actually an official method.
What about gsynk or freesynk. Do they work on here?