Tommu
patreon
Quick Update
🕑 Added 2025-03-05 11:56:25 +0000 UTCI have Good news and unfortunately Bad news.
Good news
The new PC is done and I am currently transferring all my data from my previous PC to the new one.
Bad news
The GPU that I'm using, RTX 5080 is too new and the software (pytorch) does not support it yet. So I can't generate anything at the moment, I've been troubleshooting for hours trying to find the answer and I'm drawing blanks. Perhaps some of you have an idea how to fix this or a workaround to make it work somehow. I might try ComfyUI to test whether I can generate anything at all at the moment but I'm also struggling to make it work the same as Stable Diffusion Forge. Unfortunately, this will inevitably delay production and I really don't want to call it quits just yet and just wait for the official support to come out. So for the next few days and hopefully not weeks, I can't post anything just yet until I can make this work somehow on my new PC.
For those who have encountered the same problem or might have an insight to my problem, it looks something like this.
UserWarning:
NVIDIA GeForce RTX 5080 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 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
And
RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
I've been reinstalling using different version of CUDA, Python, and Pytorch to somehow make it work but it still didn't. If any of you can help me please contact me as I am not tech savvy regarding the technical side of things.
Well if nothing works then I will have to continue using my old PC until the new one works.
Update 06/03/2025 - 14:24:
I managed to get it to work but the extensions still doesn't, so I'm working on fixing or trying to make it work somehow. If anyone knows how to make Adetailer works for a 50-series card do reach me out.
Comments
D.P
Thanks again, I'll probably try to make it work tomorrow since I've been doing this for hours already.
Hybrich
Wait, but aren't there wheels for Windows?, for example, this one should be for Python3.9, Windows (AMD64 arch), from March 4th: torch-2.7.0.dev20250304+cu126-cp39-cp39-win_amd64.whl You do have to scroll down a bunch, since there are a LOT of releases, with different dates, Python and CUDA versions. I managed to install a torch 2.7 nightly release in my Windows PC, but since I don't have a 5000 series card I can't tell if it'd work with the 5080.
D.P
Nightly version only supports Linux and not Windows unfortunately. ;_;
D.P
Thanks for showing me
Hybrich
Hey, I've read that the "beta" pytorch versions (for 2.7.0) have support for Blackwell GPU (50X0 series), so you can try that if you haven't yet. In this link there's multiple wheels for this version, just pick one of the torch-2.7.0.dev versions that use your python versions: cp310-cp310-win_amd64.whl for Windows and Python 3.10, for example: https://download.pytorch.org/whl/nightly/torch/
