CUDA support for RTX 50x on Windows

Hi,

Any word when there will be full pytorch support for sm_120 on Windows? AFAIU it’s currently only on Linux? I’ve just upgraded GPU and I’m now getting a bunch of errors like this -

G:\framepack\framepack_cu126_torch26\system\python\lib\site-packages\torch\cuda_init_.py:235: UserWarning:
NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5060 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

The nightlies don’t work, because I’ve tried that. Very frustrating having a racehorse and having to ride a donkey. :slight_smile:

PyTorch supports all Blackwell architectures on Linux and Windows since its 2.7.0 release (and beforehand via nightlies) via its CUDA 12.8 builds.

Hi, thanks very much for your response. Unfortunately despite your claim, I’m still getting errors on my older pytorch apps like the Framepack one I included in my original post.

NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.

Is there a disconnect somewhere?

Thanks again.

Yes, based on the error message you are installing a PyTorch binary using older CUDA runtime dependencies and need to uninstall it and install any PyTorch binary with CUDA 12.8 by selecting it from our install matrix.