Hi,
it seems that natively on windows nvidia/windows supports GPU memory fallback options when the dedicated memory is used up -I can see this in task manager working ok.
However, when using WSL it doesn’t seem to use GPU memory only dedicated memory (unless this is the linux application limitation or just wsl/nvdia not supporting this option) - I can see it using the windows drivers and cuda versions - and using the latest nvidia-smi drivers on linux (which are different to windows) - if they don’t support it any idea why?
Linux drivers
NVIDIA-SMI 550.120 Driver Version: 566.14 CUDA Version: 12.7
Windows drivers
NVIDIA-SMI 566.14 Driver Version: 566.14 CUDA Version: 12.7
I’ve set the virtual OS to 50GB memory just incase like windows it allocates half as GPU memory buffer but no difference.