Hey everyone,
I just got an RTX 5090, and I’m running into issues using it with PyTorch.
It looks like PyTorch doesn’t yet support the new Blackwell architecture (compute capability 12.0 / sm_120).
When I try to run any CUDA ops, I get this warning and crash:
UserWarning: NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
I’m currently on the latest stable build (CUDA 12.6).
Before switching to the nightly, I wanted to ask:
-
Is there any PyTorch build (nightly or experimental) that already supports sm_120 or any plan for it?
-
Or should I wait for a future release (like PyTorch 2.5 or the CUDA 13 build)?
-
If it’s not out yet, can I build from source with CUDA 13 to get it working?
Would love to hear if anyone has managed to get their RTX 5090 / 5080 working yet.