CUDA version discrepancies


Im pretty new to PyTorch and using GPUs in my data science projects and I’m trying to learn as much I can!

Im confused why when I check my GPU’s information with nvidia-smi it shows a CUDA version of 11.4 but when I run my model for training it gives me an error saying my Quandro K4200 GPU only has a CUDA capability of 3.0 which is too old for PyTorch to run. Any idea on why I see this discrepancy? And is it possible for me to access the CUDA 11.4 that my system says it has?

Thanks for your help!

(the error I get)
Found GPU1 Quadro K4200 which is of cuda capability 3.0.
PyTorch no longer supports this GPU because it is too old.
The minimum cuda capability supported by this library is 3.7.

The CUDA version reported in nvidia-smi is related to the NVIDIA driver you have installed.
Note that the PyTorch binaries ship with their own CUDA runtime, cuDNN etc. and your local CUDA toolkit (if installed) will only be used to build PyTorch from source or custom CUDA extensions.
As the error message explains, your GPU is too old since compute capability 3.0 is not supported anymore. The current binaries support all compute capabilities from 3.7 to 9.0 (if the CUDA 11.8 runtime is used).

1 Like