Is there a way to allocate remaining GPU to your code on PyTorch?

Question

1. Is there a way to allocate the remaining memory in each GPU for your task?
2. Can I split my task across multiple GPU's?

nvidia-smi for your reference

Answer 1

1. Yes. PyTorch is able to use any remaining GPU capacity given that there is enough memory. You only need to specify which GPUs to use: https://stackoverflow.com/a/39661999/10702372
2. Yes. GPU parallelism is implemented using PyTorch's DistributedDataParallel