I'm curious if anyone here would know what the bottleneck is when loading data into VRAM from CPU RAM on AWS. Myself and my teammates are seeing surprisingly slow load times for models. Our local training hardware doesn't have this issue. Would GCP be faster?
