System Memory Fallback for Linux

Have a model that is throwing CUDA out of memory errors on my development box with a 16GB 4080. I can find information on System Memory Fallback on Windows to fallback to system memory, however I am unable to find any information on how to do this in Linux.

Any help would be appreciative.

This error is being thrown when running a python script.

Sorry, not supported by the nvidia linux driver.

Would you happen to know if this support is roadmapped for Linux? It’s pretty useful for running inference closer to device maximums without worrying so much about killing the job.