3 ms·
Yep, I only use open source tech and forgot that people sometimes have closed source dependencies haha. Although I use Nvidia GPUs and in general the support fo
by jarbus 3y ago
Yep, I only use open source tech and forgot that people sometimes have closed source dependencies haha. Although I use Nvidia GPUs and in general the support for their drivers are still way better than AMD’s which are more open. But aside from that, the article assumes that you are using open source tech.
- gchamonlive 3y agoThis is the way!
- jchw 3y ago> Although I use Nvidia GPUs and in general the support for their drivers are still way better than AMD’s which are more open. On open source desktops, my general experience with AMD GPUs has been very positive, with the main caveat being that I don't buy AMD GPUs near product launch (which, right now, is probably not a bad decision for just about anyone, since the new products are rarely a particularly good deal). I used NVIDIA GPUs on Linux from like 2004-2016 or so, and things have just changed a lot in the intervening years in my point of view. There used to be all kinds of issues, but it has started to go the other direction, and now it's getting harder to get NVIDIA drivers working right sometimes. When people ask me personally, I recommend AMD GPUs on Linux to basically anyone that isn't actively interested in using CUDA.
- jarbus 3y agoUnfortunately my main use case is CUDA, haha. I’ve got hope that long term ROCm can catch up.
- jchw 3y agoOh yeah, same here. I'm not holding my breath, but I'll tell you what I'd really like: I'd really like to see Vulkan become the lingua franca of GPGPU compute. It's probably more possible for something like that to happen now than ever, since I think the majority of GPGPU usage is ML via PyTorch. That said, it's pretty obviously very different from CUDA, it feels unclear if this is really a promising path to go down.
- jarbus 3y agoThat would be epic, never even considered this possibility