The official page of ROCm/PyTorch will contain information that is always confusing. On this page we will endeavor to describe accurate information based on the knowledge gained by GPUEater infrastructure development.
It seems AMD is catching up with Nvidia. I am thinking about considering AMD GPUs instead of Nvidia because of the scarcity of 30 series.
Need Some infos.
If you don't want to waste a lot of time trying to get ROCm running in a configuration you want, stay with NVIDIA ... (even getting standard DL frameworks running is a pain)
It seems AMD is catching up with Nvidia. I am thinking about considering AMD GPUs instead of Nvidia because of the scarcity of 30 series. Need Some infos.