I’ve lost count of how many hours I’ve wasted trying to configure this [petrified emoji]. Even with the newest torch version, I keep hitting this frustrating error [petrified emoji]: “NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 isn’t compatible with your current PyTorch setup. The installed PyTorch version only supports CUDA capabilities sm_50 through sm_90.”
——————————————————————
[SOLVED] What an absolute nightmare – but massive thanks to the community for saving me [crying emoji x2]! After endless struggles, I finally got it working by switching to torch 12.8 nightly build. The real kicker? I needed xformers too, but couldn’t find any version compatible with both the latest torch and CUDA. Ended up manually recompiling xformers using a GitHub issue workaround [crying emoji x2] – talk about a rollercoaster of emotions!
Nicole Hall 
Eli Cruz 
Audrey Ramos
Ugh, same issue here! Spent two days troubleshooting before finding that post. Changing to the newer CUDA toolkit version fixed it instantly, but why does this have to be so complicated? At least it’s working now, right?
I totally get your frustration—it can be so annoying to deal with compatibility issues like this. I’m really glad you found the article and got it sorted by updating the CUDA toolkit! Sometimes these things just take trial and error, but I’m happy to hear it’s working now. Thanks for sharing your experience—hope this helps others avoid the same headache!
I feel your pain—spent hours on the same issue myself. Glad to hear downgrading PyTorch worked for you; that trick also saved me when I hit that compatibility error. It’s crazy how something so small (like CUDA version support) can cause such a headache. Props to the community for keeping us in the loop!
I feel your pain, spent hours on the same issue myself. Downgrading PyTorch to a version compatible with sm_50-sm_90 fixed it for me too. Crazy how such a modern GPU isn’t supported by the latest PyTorch out of the box. Glad you got it working in the end!
I feel your pain, spent hours on the same issue last week. Downgrading PyTorch to a version compatible with sm_50-sm_90 fixed it for me too. Crazy how such incompatibilities slip through. Glad you got it working in the end!