rahul003 commented on issue #9629: do not save gpu memory during fp16 training URL: https://github.com/apache/incubator-mxnet/issues/9629#issuecomment-382485782 @tornadomeet / @315386775 Were you both using such low batch sizes? I imagine in such cases, the CUDA overhead might dominate network GPU costs and does not provide a fair picture. I've tried using larger batch sizes for the same model and did notice decrease in memory used.
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
