All geek questions in one place

How can I use intel-mkl with tensor flow

I saw a lot of documentation about using a tensor stream processor, however I don't have a GPU. What I have is a fairly powerful processor and having a 5 GB nuclear core, which I hope can help me speed up the tensor flow.

Does anyone know how I can "make" tensor flow using intel-mlk?

+4
source share
3 answers

Create TensorFlow 1.2 from the source and enable MKL support during the setup phase.

Note for Mac users

As of December 2017, MKL only works with Linux. See Tensorflow.org/performance/performance_guide#optimizing_for_ cpu

. MKL TensorFlow 1.2 Linux. , --config = cuda. ​​

+3

tensorflow Eigen, Eigen MKL, :

  • EIGEN_USE_MKL_ALL Eigen
  • MKL (. MKL)
  • 64- LP64 ( ILP64)

, - , , . , Intel, , : -O3 -xHost -ipo.

0

I know it has been a whole year, but now I see that there is a WHEEL office for Intel Optimized Tensorflow. Worth a try https://software.intel.com/en-us/articles/intel-optimized-tensorflow-wheel-now-available

0
source

Source: https://habr.com/ru/post/1655726/

More articles:


All Articles