Your 3090 needs CUDA 11.x while you’ve installed the PyTorch binaries with CUDA 10.2.
Update PyTorch to the latest stable or nightly release with CUDA 11.7 or 11.8 as described here.
nvcc --version reports the version of the CUDA compiler installed in your local CUDA toolkit and is not related to the CUDA runtime and libraries used in the PyTorch binaries.
The binaries ship with their own CUDA dependencies and your local CUDA toolkit will be used to build PyTorch from source or any custom CUDA extension, so you would still need to install the updated PyTorch binaries as described in my previous post.
If PyTorch is still unable to detect and use your GPU it would point towards a driver issue and you might need to reinstall the NVIDIA driver if needed.
@ptrblck, thank you. I tried to install a lof of time GPU driver, but I don’t know what it is the best method to do it properly. Could you please give me some advices, or process to do it for NVIDIA Get Force Rtx 3060 in Ubuntu 22.0?
@ptrblck, sorry for bothering you about this issue. I reinstall my driver for GPU (530), and my cuda (11.7), in my laptop with NVIDIA geForce RTX 3060. But my GPU is still staying unvailable. Do you have any idea of the issue?
No, I don’t know what might be causing the issue but would also recommend checking if any other application is able to use your GPU.
In the past a few users were running into issues on their laptop as they didn’t realize the laptop disabled the GPU to save power and falls back to a weak on-board video output.