HierarchicalCopyAllReduce is extremely slow

sx_f April 29, 2021, 9:15am 1

I found HierarchicalCopyAllReduce is much slower than NcclAllReduce, related issues of multi-Gpus training · Issue #971 · google/automl · GitHub. Any ideas?

1 Like

Topic		Replies	Views
Why `tf.keras.applications` is so slow? General Discussion api , keras , performance , help_request	1	873	July 7, 2021
Training speed of cnn model is too slow even after using google colab General Discussion models , gpu	2	694	November 16, 2023
Training with multi-gpus can not accelerate General Discussion gpu , help_request	2	439	December 13, 2022
Slow (2x30s) model load on VM pass-through GPU with NVLink General Discussion gpu , help_request	0	1652	November 30, 2021
Self-supervised contrastive learning with SimSiam Show and Tell keras , learning , education	10	3461	July 8, 2021