rahul003 commented on issue #9774: mx.io.ImageRecordIter does not respect dtype argument URL: https://github.com/apache/incubator-mxnet/issues/9774#issuecomment-365769322 Thanks for the explanation. Btw, is training in fp16 supposed to be ~2x faster than fp32 for a given batch size? Or is this only about reduced memory usage so we can use larger batch sizes. I have run the examples you had added for fp16 in image classification, and I see maybe +- 10-15% speed changes, nowhere close to 2x for a given batch size. Is this normal? I ran these tests on p3.8x and p3.16x EC2 machines, which use the Volta range of GPUs. I have CUDA9 as well.
---------------------------------------------------------------- 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