I am trying to run some machine learning algorithms using Tensorflow, CNTK, and MxNet. Do these frameworks have some specific performance analysis tools especially to profile runtime information ( memory, communication etc.)?
Asked
Active
Viewed 231 times
2 Answers
1
FYI: MXNet has a built-in profiler. You can find more information about it here:
https://mxnet.incubator.apache.org/how_to/perf.html#profiler
Bhavin Thaker
- 61
- 5
0
Can answer about Tensorflow only. It has embedded tracing but it can give you just some basic performance characteristics. However, there is a way how you can import its tracing data to Intel VTune Amplifier (the tool is not free, but there are free fully functional academic and trial versions) and get more advanced metrics (utilization of various blocks, bandwidths between them, IO, etc) mapped to tensors.
Alexandr Kurylev
- 104
- 2