Dear Users,
I am new to the community. This is my first post since I could not figure out a way out of an issue I met lately.
I work with a Windows 10 machine with an Nvidia P2000 mobile dedicated GPU to run deep learning models. Visual Studio code, Python 3.10, Tensorflow (and Keras) 2.9, Cuda Toolkit 11.2 and cuDNN 8 are installed and with that setup, Tensorflow ran the models on the GPU automatically. No problem whatsoever.
Since two weeks ago, the computer (Dell command update) all model run on the CPU, no matter what I try to do.
My code, Tensorflow, Python and hardware are the same, so I thought that the problem must lie between the GPU and my code, i.e. the GPU drivers and when I run an ‘nvidia-smi’ command the answer is a CUDA Version 12.0, while before it was 11.6. So I went to check the drivers and I found out that they were the latest ones and had been installed at the end of January (version 528).
I tought that this mismatch between the CUDA drivers and the CUDA toolikt could be the culprit and thus I downloaded the 460.89 driver version (whose CUDA version is 11.6) and (clean) installed them.
After rebooting, nvidia-smi tells me that the CUDA Version is 11.6, but my code still runs on the CPU.
I would like to ask if someone else found this issue and how it was solved.
Thanks in advance!!