[GitHub] cjolivier01 commented on a change in pull request #8553: Engine reserves cores from OMP. Set some defaults for dynamic and recursion

[GitBox <gi...@apache.org>]

cjolivier01 commented on a change in pull request #8553: Engine reserves cores from OMP.  Set some defaults for dynamic and recursion
URL: https://github.com/apache/incubator-mxnet/pull/8553#discussion_r148977213
 
 

 ##########
 File path: src/engine/threaded_engine_perdevice.cc
 ##########
 @@ -234,6 +234,27 @@ class ThreadedEnginePerDevice : public ThreadedEngine {
     }
   }
 
+  /*!
+   * \brief Get number of cores this engine should reserve for its own use
+   * \param using_gpu Whether there is GPU usage
+   * \return number of cores that this engine wishes to be reserved
+   * \note Testing found no degradation of performance using these values
+   *       running cifar10 with resnet50 on various GPU systems,
+   *       including AWS p2.16xlarge, which has 16 GPU's
+   */
+  int GetReserveCoreCount(const bool using_gpu) const {
+    int reserve = 0;
+    if (using_gpu) {
+      // Save at least one for GPU tasks
+      ++reserve;
+      // If we have 8 or more real cores, reserve another core for GPU tasks
 
 Review comment:
   I tried cifar10 with res50 on a p2.16xlarge and it had same images/sec as master branch. If you want a couple of extra reserved -- at least in the short term -- that's fine.

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services