You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@mxnet.apache.org by "Da Zheng (JIRA)" <ji...@apache.org> on 2018/03/14 20:57:00 UTC

[jira] [Created] (MXNET-105) Fix CuDNN performance after code refactor

Da Zheng created MXNET-105:
------------------------------

             Summary: Fix CuDNN performance after code refactor
                 Key: MXNET-105
                 URL: https://issues.apache.org/jira/browse/MXNET-105
             Project: Apache MXNet
          Issue Type: Improvement
            Reporter: Da Zheng


We observed about 4% performance decrease reported inĀ [#9874|https://github.com/apache/incubator-mxnet/issues/9874]. There are multiple factors that cause the performance decrease.
 * The first one is that the refactored code passes more arrays to the backward of BatchNorm.
 * The second one is that the refactored code needs to reinitialize the CuDNN states in every forward and backward.
 * The third one is that the refactor code leads to more memory allocation (e.g., creating std::vector) in forward and backward. (However, the test shows that this doesn't cause much performance decrease.)



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org
For additional commands, e-mail: issues-help@mxnet.apache.org