zhaoningning commented on issue #7414: why rnn train speed is not 
stable,sometims very slow?
   I used mxnet.callback.Speedometer, so yes, I think 30 samples/sec is the 
mean speed  for a batch,and I fixed the input image size, so it's not variable 
sequence length;  
   ps: I tried training two models at the same time(using the same cnn+rnn ) 
with different gpus on a server,the training  speeds of two process almost the 
same,and when one training speed become 600 samples/sec, the other training 
speed also grow to nearly 600 samples/sec, when one training speed slow 
down,the other training slow down  as well;  And  during the low training speed 
time, most of cpus is  free, and gpus are  almost free.
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:

With regards,
Apache Git Services

Reply via email to