bill10 commented on issue #2919: URL: https://github.com/apache/incubator-mxnet/issues/2919#issuecomment-789167389
Found this old issue, and hopefully it is resolved? I encountered a similar problem: when I trained a model on a P2.8x on AWS it was all fine but when I moved to P2.16x to take advantage of data parallelism it got GPU memory error. I was using kvstore = 'device' and update_on_kvstore = False. It seems that the kvstore device is trying to copy gradients from all other GPU's at once and use up all memory, as described in this issue. Any thoughts? Thanks! ---------------------------------------------------------------- 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: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
