cyrusbehr opened a new issue #19729:
URL: https://github.com/apache/incubator-mxnet/issues/19729
When I run GPU inference, I get the following message:
```
[21:47:34] ../src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running
performance tests to find the best convolution algorithm, this can take a
while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to
disable)
```
This significantly slows things down as tuning can take some time, and is an
issue when I have variable input sizes (as tuning can occur for every image). I
therefore want to disable tuning.
If I run `export MXNET_CUDNN_AUTOTUNE_DEFAULT=0` then re-run my script, it
makes no difference. The message is still printed, and auto tuning is still
performed. Why is the environment variable not being used?
I am building mxnet release 1.6.0 from source, using the following cmake
flags:
```
cmake\
-DBLAS=open\
-DUSE_CUDA=ON\
-DUSE_CUDNN=ON\
-DMXNET_CUDA_ARCH="6.0;6.1;7.0;7.5"\
-DENABLE_CUDA_RTC=OFF\
-DCMAKE_BUILD_TYPE=Release\
-DUSE_F16C=OFF\
-GNinja\
-DUSE_LAPACK=OFF\
-DUSE_LAPACK=OFF\
-DUSE_JEMALLOC=OFF\
-DUSE_CPP_PACKAGE=ON\
-DUSE_SIGNAL_HANDLER=OFF\
-DUSE_OPENCV=OFF\
-DUSE_MKL_IF_AVAILABLE=OFF\
-DUSE_MKLDNN=OFF\
-DBUILD_CPP_EXAMPLES=OFF\
-DCMAKE_INSTALL_PREFIX=./packaged\
..
```
Using 18.04 and Cuda 10.1.
----------------------------------------------------------------
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:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
