Announcing the Intel® Distribution for Python* Release

I am pleased to announce the first product release of Intel® Distribution for Python* powered by Anaconda*. Why is Intel making a python distribution? We want python users to get the productivity they are accustomed to without having to compromise on performance. Our customers told us that it was difficult to get a Python that could fully utilize the latest processors. A single socket Xeon Phi system can do 4600 single precision FLOPS/cycle, but some of the commonly used python software stacks only could do 4 FLOPS/cycle because they did not leverage multiple cores or the latest SIMD instructions.

Intel Distribution for Python contains a python interpreter and highly optimized versions of the core numeric packages. We use  Intel® Math Kernel Library (Intel MKL), Intel ® Threading Building Blocks (Intel TBB),  and the Intel® Compiler so whether you  are using an Atom, Xeon, or Xeon Phi processor you will get great performance for computationally and data intensive programs.

We know performance is not everything. You are likely using Python because of the large number of high quality packages. We wanted to ensure that you would have easy access to the commonly used python packages. We have been working closely with Continuum Analytics and are now proud to say Intel Distribution for Python is “Powered by Anaconda.” We use the same build recipes as Anaconda to ensure high compatibility between the distributions. We use conda for package management so you have full access to the large and growing set of conda packages, and we publish our packages on Anaconda Cloud so Anaconda users are only a few keystrokes away from trying out Intel Distribution for Python in a familiar environment.

Our measurements show that python can get close to native performance for the key algorithms in dense linear algebra, but there is room for improvement in many of the algorithms in numpy, scipy, and scikit-learn. Our python distribution delivers large speedups for threading, random number generation, and FFT. We are working to contribute these improvements upstream so all python users will benefit, whatever way you obtain python.

Please try out Intel Distribution for Python and let us know what you think.

Robert Cohn

For more complete information about compiler optimizations, see our Optimization Notice.

1 comment


Add a Comment

Have a technical question? Visit our forums. Have site or software product issues? Contact support.