Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[REQUEST]: Support for Blackwell GPU (50xx) #1464

Open
samw1tch opened this issue Feb 23, 2025 · 7 comments
Open

[REQUEST]: Support for Blackwell GPU (50xx) #1464

samw1tch opened this issue Feb 23, 2025 · 7 comments

Comments

@samw1tch
Copy link

samw1tch commented Feb 23, 2025

In a few words, describe your idea

Please make a version with support for pytorch 12.8

More information

Recently got a 50xx series card and we have no support for it. I'm too much of a skiddy to update the install myself. Help would be appreciated!

Image

@Kuuko-fokkusugaru
Copy link

That's weird. Could you specify the specs of your computer, os, and RVC version?

@samw1tch
Copy link
Author

Ah, sure. I have a 7800x3d, 5070ti, and the newest version of drivers etc. I'm running on win11, and I'm also having this issue with the latest version of the CUDA windows release. I'd venv and fix it, but again, very low level of knowledge when it comes to python, haha.

Image

Image

@samw1tch
Copy link
Author

of course, the solution is to just pip install torch 128 nightly into the location, but a prebuilt installer would be better for those like myself to one click go.

Image

@Kuuko-fokkusugaru
Copy link

Out of curiosity, I guess this issue also affects v1? v1 for many, if not all, performs better in terms of quality. Except for a few users, everyone always ends up going back to v1 for that very reason.

@samw1tch
Copy link
Author

Yup, and the problem is that we don't have a version of torch that's an official release, and other AI related apps had to have theirs built beforehand. I expect this to be eventually resolved in like, two months.

@Kuuko-fokkusugaru
Copy link

would that manual fix for v2 work for v1 too? I feel it may be quite important for many to be able to fix v1 while it outperforms v2

@samw1tch
Copy link
Author

Not yet. I've officially tried everything including building my own version of torchaudio for 12.8, but as of 2/25, there's still no version of torchaudio specifically that will work with cuda 12.8. someone is gonna have to build it themself, and unfortunately I am not nearly as skilled enough to do so,

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants