5 ms·
Unfortunately AMD doesn't really compete in the machine learning space. Hopefully one day ROCm will offer a viable alternative but until then we are pretty much
by fungi 6y ago
Unfortunately AMD doesn't really compete in the machine learning space. Hopefully one day ROCm will offer a viable alternative but until then we are pretty much captive.
- vetinari 6y agoIf you really need nvidia so badly, put that GPU into headless machine, where you don't need any UI. Otherwise, you will be frustrated on desktop and it still won't work out of the box.
- beojan 6y agoOr just use integrated graphics for the desktop, and reserve the GPU for GPGPU work.
- vetinari 6y agoYes, that would be an option too.
- Havoc 6y ago>put that GPU into headless machine Possibly one of the saddest solutions ever proposed. Besides then one can’t game on it
- vetinari 6y agoThe original argument was the need for CUDA. It works headless, no sadness about that. What is sad, is to reward Nvidia with your money despite being full aware, that they don't want you to have a nice Linux experience as the other vendors do. Muh games fps is just a nail in the coffin.
- disgruntledphd2 6y agoNvidia do care about Linux. Almost all of their cloud/AI sales run Linux. So CUDA normally works, but the graphics drivers themselves are crap. If I were to specify a system most likely to "just work" with Nvidia, I'd probably go for whatever version of CentOS FB, Google and Amazon use internally.
- vetinari 6y agoAs you wrote, they might care about their cloud and AI sales, but for linux desktop and OOTB experience? If they do, they hide it very well.
- washadjeffmad 6y agoWith VFIO on KVM, you can. You can make your own Stadia with actual games that way with a raspi running SteamLink.