Using an nVidia Tesla p40 for rendering on Linux - eviltoast

I’ve seen it covered widely for Windows, but I don’t know if people have gotten it working on Linux.

I’ve been thinking about buying an nVidia Tesla p40 off of eBay for a GPU upgrade. Currently I’m running a Quadro M2000 in my Dell Precision T7910 with dual Xeon E5-2620 v3 processors. Obviously, I’ll have to work out how to cool it but apparently people have had success with GTX 1080 coolers for dead graphics cards.

I’ll need to keep the Quadro for video output since the Xeons have no integrated graphics. I’m hoping it would behave like an Optimus laptop, so I don’t have flickering in Xwayland (in preparation for Fedora 40, plus I prefer Wayland anyway).

If anyone has attempted this maybe with another Tesla card I’d like to know how it went.

  • RedBauble@sh.itjust.works
    3·
    9 months ago

    I used to do this with a nvidia tesla m40 and a radeon hd6850. Used the tesla for rendering amd encoding, the radeon for display output. I just followed the arch wiki pages related to nvidia optimus laptops and PRIME offloading. It worked but was a bit junk, in some other tests I did, when the radeon was used to render the DE, I had a much more fluid experience, offloading the rendering seems to lead to some micro stutters every now and then that make it a not so fluid experience. But ymmv I guess. Also I haven’t had any luck with two separate nvidia cards, but that was probably due to driver version mismatch between the two cards

    • ⸻ Ban DHMO 🇦🇺 ⸻@aussie.zoneOPEnglish
      1·
      9 months ago

      Thanks! Good to hear its like a laptop since I’ve been using optimus laptops for a while now as well.

      What did you do to keep the card cool? …And was it loud?

      • RedBauble@sh.itjust.works
        21·
        9 months ago

        What did you do to keep the card cool?

        Poorly. Had 3d printer a fan duct and ducted a fan to the back of the case, to push-pull air. Those cards are made to work in server racks, with really high pressure and high speed fans, not really for a desktop. I have seen people on reddit mounting a modified 3070ti cooler on the tesla, but I had not had a chanve to try that.

        And was it loud?

        Yes, depending on the fans used. But high speed fans are generally loud. Also lots of vibrations, but that qas mostly fault of my incredibly sketchy setup

      • just_another_person@lemmy.world
        3·
        9 months ago

        Redirecting output aside, these are datacenter cards, and are scaled differently than gaming cards. Lower clock speeds and such. I’m not sure how good for actual gaming they would be.

        • ⸻ Ban DHMO 🇦🇺 ⸻@aussie.zoneOPEnglish
          1·
          9 months ago

          There’s a few factors that drew me towards the Tesla card. The main one being size, because of the way the precision is designed there’s really no room for tall graphics cards with big coolers and a power connector on the side. I’ve been looking at the Quadros as well but I think they’re still a bit expensive for the performance and age. Gaming isn’t the only purpose this card would get use for either, probably some Cuda/OpenCL stuff, maybe play around with some AI stuff and power a Windows VM (vgpu). At the moment the Tesla P40 is around $280AUD ($182USD) while the Quadro P4000 is around $400AUD ($261USD). An RTX 3050 8gb would probably be alright but there don’t seem to be any in the second hand market for a reasonable price

          • Nomecks@lemmy.ca
            2·
            9 months ago

            So these cards have only heatsinks and no fans. They’re designed to be used in hot/cold aisle facilities with the server cooling package moving the air. You’re going to need to work around that if you want to use one, or you’ll quickly overheat the card.

          • fuckwit_mcbumcrumble@lemmy.worldEnglish
            2·
            9 months ago

            There’s kind of a reason why they’re so cheap. They’re fiddly to deal with, and take a lot of coercion if you want to make it do anything other that what it was designed for (be a headless rendering farm for videos and maybe AI). That said I did find this which might do what you want it to do. https://www.reddit.com/r/linux_gaming/comments/slzc2u/shocked_and_impressed_by_the_state_of_hybrid/

            I want to say back in the Haswell days there was a weird push for having hybrid GPU setups on desktops, and the iGPU would actually help the dGPU in some cases. Maybe you could find something from around then to coax that GPU into working?

            That said do you need a Quadro or god forbid the Tesla for what you’re doing? A regular ass GTX 1080 would probably better for what you’re doing. Founders edition GPUs are a conventional blower and are only the height of the pcie slot (plus power cables).

            • ⸻ Ban DHMO 🇦🇺 ⸻@aussie.zoneOPEnglish
              1·
              9 months ago

              Thanks, I might just start with an RX 580 since they’re quite cheap here and keep a Quadro around for CUDA stuff and upgrade that to a Tesla in the future if needed.

              The most graphically demanding game I have is Age Of Empires III Definitive Edition and I only have 1920x1080 monitors. It’s actually playable at near max settings on the Quadro but wouldn’t mind a frame rate boost. I’ll probably just need to remove the support bracket on the inside of the side panel