2 ms·
Microsoft has DirectML and a PyTorch backend for it. Although it is stale and significantly slower on 3090 than CUDA (5x+ as of 2 years ago).
by lostmsu 3y ago
Microsoft has DirectML and a PyTorch backend for it. Although it is stale and significantly slower on 3090 than CUDA (5x+ as of 2 years ago).