ptrendx commented on issue #2919: URL: https://github.com/apache/incubator-mxnet/issues/2919#issuecomment-789246602
@bill10 Probably the most efficient way of doing data parallel training currently is using Horovod. You should be also able to use kvstore = 'nccl', but I believe that would be slower. A side note - why using P2 instances? The P3 (not to mention P4) would be significantly faster (though more expensive), and in fact even G4 (using Tesla T4 GPU) should be at least comparable if not faster (especially in mixed precision where it should be much faster than P2) while being cheaper. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
