apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-574402879
No regression found when building mxnet from source with 1.4, 1.5 and 1.6.
The difference in performance could be due to the in
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-574029399
The original issue was regression between 1.4 and 1.5. Could you also post
results built from source using 1.4?
--
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-572184919
@jonatan1626 Could you please provide an update? What's the final conclusion
or action items for the performance from 1.4 to 1.
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-57629
@jonatan1626
When we say test pip wheel, we mean using the pip wheel published at PyPi.
When build from source, we always u
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-569968772
@jonatan1626 When building from source please use exactly the same cmake
command as @wuxun-zhang is using. The ./build.sh is us
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-569807927
@jonatan1626 Yes. That's an apple-to-apple comparison as we discussed.
Can you also verify if there is degradation betwee
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-569155617
@wuxun-zhang Based on your earlier performance result, it seems we still see
inference runtime increased from Avg: 14.64 in mxn
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-569140957
> EIA context not available, trying GPU...
Why is this line ever printed at all? Apache mxnet should have no knowledge
o
apeforest commented on issue #17159: Performance regression from 1.4.1 to 1.5.1
URL:
https://github.com/apache/incubator-mxnet/issues/17159#issuecomment-568779536
@pengzhao-intel performance in 1.6 came back a little but still not as good
as 1.4. Also we are not sure if the performance rec
