4 ms·

Stable Diffusion works for me with a Polaris GPU. Had to compile my own local copy of Tensorflow to use it, but everything runs.

by pja 4y ago

cmxch 4y ago
Which documentation/build environment are you using? I’m using Ubuntu(to follow what AMD has for ROCm) and building the entirety of (gfx803 patched) ROCm from source. It works with some forks but not others.
- MayeulC 4y ago
  I am in the same boat with a gfx03 card. What patch did you use? The ones here? https://github.com/xuhuisheng/rocm-build https://github.com/xuhuisheng/rocm-build I also tried to compile pytorch with its Vulkan backend, but ended throwing the towel as LDFLAGS are a mess to get right (I successfully compiled it, but that was only part of the build chain, and decided I had better things to spend time on). I wonder how that would perform; ncnn works pretty decently.
  - pja 4y ago
    I did try that, but I don’t think the pytorch Vulkan backend is complete enough. I never did get a working install going down that route though, so I could be wrong.
  - cmxch 4y ago
    Yes. I went through their build instruction shell scripts with some issues but it did work as below: It works with an older fork that has img2img / txt2img separated from each other, but nothing else did. When you built it, you did at least source the env.sh to pick up the build environment that it needed?
- pja 4y ago
  IIRC it’s just the standard AMD build from https://repo.radeon.com/rocm/apt/5.2.3/ https://repo.radeon.com/rocm/apt/5.2.3/ I think. It’s possible I had to do something weird, but I think it supports Polaris OOB. rocminfo certainly thinks so. I’m running it on Debian testing, with an equivs package to get it to install cleanly: $ cat rocm-equivs Package: amdgpu-driver-fixes Provides: python,libstdc++-7-dev,libgcc-7-dev Architecture: all Description: Fixes the AMD GPU driver installation on Debian testing I had to compile my own version of pytorch to get gfx803 support there. I‘ll see if I can recreate the steps & create a runbook.

Stable Diffusion works for me with a Polaris GPU. Had to compile my own local copy of Tensorflow to use it, but everything runs.

by pja 4y ago

Stable Diffusion works for me with a Polaris GPU. Had to compile my own local copy of Tensorflow to use it, but everything runs.

cmxch 4y ago

Which documentation/build environment are you using? I’m using Ubuntu(to follow what AMD has for ROCm) and building the entirety of (gfx803 patched) ROCm from source. It works with some forks but not others.

MayeulC 4y ago
I am in the same boat with a gfx03 card. What patch did you use? The ones here? https://github.com/xuhuisheng/rocm-build https://github.com/xuhuisheng/rocm-build I also tried to compile pytorch with its Vulkan backend, but ended throwing the towel as LDFLAGS are a mess to get right (I successfully compiled it, but that was only part of the build chain, and decided I had better things to spend time on). I wonder how that would perform; ncnn works pretty decently.
- pja 4y ago
  I did try that, but I don’t think the pytorch Vulkan backend is complete enough. I never did get a working install going down that route though, so I could be wrong.
- cmxch 4y ago
  Yes. I went through their build instruction shell scripts with some issues but it did work as below: It works with an older fork that has img2img / txt2img separated from each other, but nothing else did. When you built it, you did at least source the env.sh to pick up the build environment that it needed?
pja 4y ago
IIRC it’s just the standard AMD build from https://repo.radeon.com/rocm/apt/5.2.3/ https://repo.radeon.com/rocm/apt/5.2.3/ I think. It’s possible I had to do something weird, but I think it supports Polaris OOB. rocminfo certainly thinks so. I’m running it on Debian testing, with an equivs package to get it to install cleanly: $ cat rocm-equivs Package: amdgpu-driver-fixes Provides: python,libstdc++-7-dev,libgcc-7-dev Architecture: all Description: Fixes the AMD GPU driver installation on Debian testing I had to compile my own version of pytorch to get gfx803 support there. I‘ll see if I can recreate the steps & create a runbook.