May I ask when CUDA 13.0 is expected to be released

vision
1

The current CUDA version of the computer is 13.0, and Pytorch 12.9 cannot be used

2

Your locally installed CUDA toolkit won’t be used as our Pytorch binaries ship with all needed CUDA runtime dependencies. You would need to properly install an NVIDIA driver before installing our binaries.

I’m any case, the CUDA 13 nightly binaries are WIP already.

3

Thanks for your reply

4

Any new developments on when PyTorch will be ready for CUDA 13?

5

Nightly binaries are already available.

1 Like
6

Hi, I’m looking on the nightly page for torch cu130. I don’t see v2.8.0 however.

Will there be a release for torch 2.8.0+cu130?

Unfortunately I have a development issue b/c I’m running CachyOS (Arch Linux). The rolling distro upgraded us to CUDA 13.0. So I’m having issues with some libraries that expect a matching CUDA version…which are stuck on 12.8. I could upgrade to the next version, but some of the builds aren’t available yet…like xformers…Though hopefully it will be out soon!

7

No, PyTorch 2.8.0 won’t be re-released. Our nightly binaries as well as the upcoming 2.9.0 release support CUDA 13.

1 Like
8

FWIW, It looks like there’s an Arch package for torch 2.8.0 built with CUDA 13. But I don’t see it anywhere in PIP.