What kept GPUs waiting: an E1 study of distributed training overhead

(nodeox.com)

1 points | by dvoronko 11 hours ago ago