Hi, after a long break from using ROCM I tried to use PyTorch with my Radeon 7600, but it doesn't get detected.
This particular GPU is not on AMD support list, but on the internet people are saying it works just fine.
When I do:
import torch
print(torch.cuda.is_available())
print(torch.version.hip)
I get false and None.
I tried doing HSA_OVERRIDE_GFX_VERSION=11.0.0 and
export ROCM_PATH=/usr
export HIP_PATH=$ROCM_PATH
export ROCM_SOURCE_DIR=$ROCM_PATH
export HIP_COMPILER=clang
export HIP_RUNTIME=amd
export HIP_ROCCLR_HOME=/usr/lib64/llvm-rocm
export HIP_CLANG_PATH=/usr/lib64/llvm-rocm/bin
export HIP_CXX_COMPILER=/usr/bin/hipcc
export HIP_DEVICE_LIB_PATH="/usr/lib64/amdgcn/bitcode"
export DEVICE_LIB_PATH=$HIP_DEVICE_LIB_PATH
but none of it seems to help.
Is anyone currently running PyTorch or something based on it like StableDiffiusion on AMD and did you have to take any additional steps to make it work?
