Esta é a mensagem recebida da execução de um script para verificar se o Tensorflow está funcionando:
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcurand.so.8.0 locally
W tensorflow/core/platform/cpu_feature_guard.cc:95] 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:95] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:910] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
Notei que ele mencionou SSE4.2 e AVX,
- O que são SSE4.2 e AVX?
- Como esses SSE4.2 e AVX melhoram os cálculos da CPU para tarefas do Tensorflow.
- Como fazer o Tensorflow compilar usando as duas bibliotecas?
NOTE on gcc 5 or later: the binary pip packages available on the TensorFlow website are built with gcc 4, which uses the older ABI. To make your build compatible with the older ABI, you need to add --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" to your bazel build command. ABI compatibility allows custom ops built against the TensorFlow pip package to continue to work against your built package.
de aqui tensorflow.org/install/install_sources
bazel build -c opt --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --config=cuda -k //tensorflow/tools/pip_package:build_pip_package
Em Xeon E5 v3 que me dá 3x melhoria na 8k velocidade da CPU matmul comparação com o lançamento oficial (0,35 -> 1,05 T ops / seg)