and in advance for your attention,
I just installed shadoworflow (on a new machine with Ubuntu 16.04 and CUDA 8.0 already installed) using the following procedure :
At first I used --copt=-march=native. I got a message
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use SSE3 instructions, but they are available on your computer and can speed up CPU calculations.
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use SSE4.1 instructions, but they are available on your computer and can speed up CPU calculations.
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use SSE4.2 instructions, but they are available on your computer and can speed up CPU calculations.
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use the AVX instructions, but they are available on your computer and can speed up CPU calculations.
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use the AVX2 instructions, but they are available on your computer and can speed up CPU calculations.
W tensorflow / core / platform / cpu_feature_guard.cc: 45] The TensorFlow library has not been compiled to use FMA instructions, but they are available on your computer and can speed up CPU calculations.
, ,
Tensorflow SSE4.2 AVX?
nVidia,
bazel build -c opt --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --copt=-msse4.2 --config=cuda -k
, . , , ?
!