I recently rented a p2.8xlarge option on Amazon's AWS EC2 to compute larger models. Unfortunately, I couldn't increase the size of my model due to repeating 'out of memory' errors. When I checked GPU memory usage via nvidia-smi, it showed me that the model was running on only one GPU. The 7 others were simply doing nothing. How can I use these?
I recently rented a p2.8xlarge option on Amazon's AWS EC2 to compute larger models. Unfortunately, I couldn't increase the size of my model due to repeating 'out of memory' errors. When I checked GPU memory usage via nvidia-smi, it showed me that the model was running on only one GPU. The 7 others were simply doing nothing. How can I use these?