The TensorFlow library wasn't compiled to use SSE3, SSE4.1, SSE4.2, AVX on Google Cloud Platform Console

Question

I have a TensorFlow model to test a wide-n-deep neural network, but I can't get it to run on my windows machine because of a bug in the TensorFlow Library. Now I'm having to resort to Google Cloud Platform. I got everything set up where my python file processes input, but when I run the code through the console, I get the following messages:

$ python -m widendeep.py -h
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.

These messages don't show when run on Windows. Is this because I can't run this through the cloud this way? Do I have to use gcloud ml-engine local train or gcloud ml-engine jobs submit training my_job? Any guidance on the proper way to do this would be great.

Answer 1

These messages don't show when run on Windows. Is this because I can't run this through the cloud this way?

You can run your code as-is, or at least if you can't it has nothing to do with the warnings you listed.

These warnings are saying that you might get better performance from TensorFlow with the machine architecture you are running it on because it supports a more advanced instruction set than the one it was compiled with.

Do I have to use gcloud ml-engine local train or gcloud ml-engine jobs submit training my_job?

I'm not very familiar with Google cloud (using Amazon at the moment myself), but I can say that if you do need to use the above commands it has nothing to do with the warnings you listed above.

These warnings apply to the CPU instruction set and to how it interacts with the GPU so the performance improvement may or may not be significant (or exist at all), depending on your specific application.

If you want to be sure you are using the full potential of the hardware you are running your program on you will need to compile TensorFlow on the platform you are running in on (check How to compile Tensorflow with SSE4.2 and AVX instructions?).