[GitHub] SumNeuron commented on issue #8751: Distributed Training has inverse results when imported (8 GPUS is slower than 1!)

[GitBox <gi...@apache.org>]

SumNeuron commented on issue #8751: Distributed Training has inverse results when imported (8 GPUS is slower than 1!)
URL: https://github.com/apache/incubator-mxnet/issues/8751#issuecomment-346143311
 
 
   @piiswrong I agree that MNIST is small, but **1.)** that does not explain that in functionally equivalent code why using 8 gpus goes from 2.5 seconds (when everything is in the same file) to 17 seconds (when the net is imported) and **2.)** I tried AlexNet and opened an issues where the DataLoader causes waaaaaaaaaaaay to much memory to be alloced. (It throws cudaMalloc errors on GPUs with 16GB of ram)

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services