medium10 pts · first answer countsgpu-computingdata-loadingstorage

Training data on a slow network share

Each epoch reads 3 TB of images from a network file share at 500 MB/s, and GPUs wait most of the time. What is the most effective fix?

A.Buy faster GPUs for the same nodesB.Copy or cache the data on local NVMe before trainingC.Increase the learning rate to finish soonerD.Lower the image batch size to 1

Discussion 0

Join free to post in the discussion.

Training data on a slow network share

Discussion 0

Related questions