Quick clarification about torch.matmul

1

Is it possible to do a matmul across a matrix divided over many gpus? Or really not?