Software Archive
Read-only legacy content
17061 Discussions

KMP, OMP, MKL configuration IntelPython 2019 with TensorFlow 1.6


I am running on a Supermicro K1SPE motherboard (BIOS 2.0a) with a XeonPhi 7250,  214Gb memory and CentOS 7.5 (pls see attached). This system is running very slow.  I originally thought it was a Jupyter NotebookChrome problem but the problem persists when running on IntelPython 2019.  I have tried many combinations of OMP, KMP and MKL to no avail.  I included the environment code from a tensorflow model which takes about 10 min to run on the XeonPhi and 2 min to run on a i5 Surface (a new Surface would have been a lot cheaper).  When I run VTune, I am only utilizing 4.65 logical CPUs out of a 272 total and memory is stalled.  I thought this would improve with IntelPython 2019 but things are just as bad. At least I can run TensorFlow 1.6 on IntelPython 2019 versus 1.3 on 2018, however the ent version of TensorFlow is 1.10.  I am sure I am not the only person with this problem as TensorFlow is a very popular application.  Please see attached scan for additional documents.  I can be reached in my office ***.***.3611 cell ***.***.7001 if you need additional information.  Phil

0 Kudos
0 Replies