hfutxrg commented on issue #7778: float16 has no performance improvement URL: https://github.com/apache/incubator-mxnet/issues/7778#issuecomment-327831887 Hi @ptrendx, yes I am using Volta GPUs. I just tried 1 V100 and found FP16 is 47% faster than FP32. So it is effective for 1 GPU. I will try larger batch size as you suggested because this improvement is still much lower than the improvement I got in Nvidia caffe. Also I will try to use larger --data-nthreads N value for multiple GPUs. Thanks for the help! ---------------------------------------------------------------- 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: [email protected]
With regards, Apache Git Services
