[GitHub] [incubator-mxnet] beew commented on issue #18879: mxnet python gpu not work

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

beew commented on issue #18879:
URL: https://github.com/apache/incubator-mxnet/issues/18879#issuecomment-670987830


   @szha 
   Hi, 
   
   1.without source /opt/intel/bin/compilervars.sh intel64, build with openblas instead of mkl (since can't find mkl root and headers) 
   python gpu examples works (from here https://mxnet.apache.org/versions/1.6/get_started/validate_mxnet.html)
   
   2 With /opt/intel/bin/compilervars.sh intel6 and build with mkl then gpu examples hang as reported here, regardless whether JEMALLOC is turned on or off.  If it is turned off, got an additional warning "Significant performance increases can be achieved by installing and enabling gperftools or jemalloc development packages"
   
   So it seems that mkl somehow doesn't play well with cuda...
   
   I haven't tried the nightly branch as it seems that the prebuild mxnet python wheels only support cuda > 10.0 after version 1.5.1,  so instead I tried pip install mxnet-cu100mkl==1.5.1  I was able to create the mxnet array in gpu context with no problem with 
   
   


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to 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