Are You Ready for a 5x Python Performance Boost?

istock 508672023

In the highly competitive world of CAD/CAE where high performance is everything, software developers can now add real value by using an accelerated Python implementation. And I have some test results from DATADVANCE, known for their products for high-end CAD/CAE model optimization, that quantify the value.

The Python language is known as being easy to learn and use. However, as an interpreted language, Python is also well known for being too slow for computationally intensive applications.

Getting higher performance with Python using accelerated Python is relatively new. Thanks to some Python aficionados at Intel, we can all use an accelerated Python that yields big returns for Python performance without requiring that we change our code. It’s painless because it’s done by switching to an updated Python that uses high-performance data analytics and math libraries underneath (including NumPy, SciPy, scikit-learn, pandas, Jupyter, matplotlib, mpi4py, etc.). Accelerated Python aims to deliver the speed of compiled languages, with full optimization for a wide range of processors, but without any code changes or compromises on compatibility.

As an example of the benefits of accelerated Python, there are some encouraging results from DATADVANCE highlighted in this short whitepaper

A notable capability of their software is the support for full “scriptability” via Python. Their reliance on Python gave them a fast path to higher performance for their customers by shifting to an accelerated Python.  Specifically, they tested the Intel® Distribution for Python* with their flagship product, the pSeven platform, and found that it boosted performance up to 5x over the standard (non-accelerated) Python distribution. No codes changes, and no changes in capabilities, happen when shifting to an accelerated Python.

DATADVANCE tested both the Intel Distribution for Python (with acceleration) and the standard Python (without acceleration). The whitepaper covering the results quoted their chief developer, Dmitry Vetrov, as saying: “We tested different version combinations and distributions of Python and NumPy for estimation of Sobol indices using pSeven Core, since it’s one of the common problems our customers solve. For older versions of Python―for example, 2.6―the boost reached even 10x, but for the newer ones it stayed around 3x to 5x.”


Free and Easy Downloads

You can learn more about (and download) the Intel Distribution for Python at software.intel.com/distribution-for-python. It’s free and it has gained considerable popularity with Python users because of its speed. The Intel packages for accelerating Python performance are also available on the Intel channel on Anaconda.org, PyPI, and YUM repositories.

The Intel Distribution for Python is also included with Intel Parallel Studio products, which feature many additional other profiling, tuning, debugging, and optimization capabilities that help programmers using Python as well as C, C++, Fortran, Java, and more.

Related Articles

·       How Does a 20X Speed-Up in Python Grab You? | InfoWorld

·       Python: High Performance or Not? You Might Be Surprised | InfoWorld

·       Why cloud platforms should invest in the promise of Python | InfoWorld

·       A more in-depth look at the specific case study cited in this article: Optimal Design with 5x Performance Boost


Copyright © 2018 IDG Communications, Inc.