Support for NVIDIA RTX 5090 (CUDA SM_120)

Hi,
I’ve recently tested PyTorch (nightly build with CUDA 12.1) on an NVIDIA RTX 5090 and received a warning:
“sm_120 is not supported”

Will support for sm_120 be included in upcoming builds?

GPU: RTX 5090

Driver: 581.42

CUDA: 12.2

One more thing!

I’m using a new NVIDIA RTX 5090 GPU (compute capability sm_120, CUDA 12.2).
When running YOLOv8 inference on GPU, I get this warning and it fails:

NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.

However, YOLOv5 runs on the same setup without issue.

PyTorch was built with CUDA 12.1 from our 2.1 to 2.5 release, which are too old too support the latst hardware generation. Blackwell requires CUDA 12.8+ so install any of our binaries built with CUDA 12.8+ instead.