Hi,
I am trying to train a model in a GPU powered machine but the GPUs do not start and all the work is done on CPUs only. The version of tensorflow of GPU is installed accordingly (1.15.4) but it does not work. The version of CUDA is downgraded to version 10 but when I run the command nvidia-msi it shows the CUDA version 11.2 but the command nvcc shows the version 10.1, does this has any thing to do with it as the wrong version might be pointed out?. Tried to uninstall the tensorflow only version (CPU one) and leave only tensorflow GPU but in this case training hangs on the beginning on initializing the variables.
Withe the same setup I am able to run the training in another less powerful machine using GPU.
Thanks in advance for the help.