MXNet Joins Apache: An Excellent Framework for Deep Learning on Intel Platforms in the AWS Cloud

Intel processors are the driving force behind many of today’s most advanced artificial intelligence (AI) applications. In the past year, Intel has added support and optimizations for many of the most popular deep learning frameworks targeting a broad spectrum of deep learning algorithms. Amazon Web Services* (AWS) and Intel have partnered together to optimize deep learning frameworks such as Apache MXNet to run best on AWS EC2 instances like the C4 and upcoming C5.

Intel is excited to see MXNet join the Apache* Software Foundation’s Incubator program and is thrilled to continue our partnership with AWS and the MXNet community. As part of this investment, Intel will continue to optimize and showcase Intel® Xeon® and Xeon PhiTM processors as a compelling platform to support deep learning applications and focus our optimization efforts to keep MXNet as one of the fastest frameworks available. MXNet is a highly scalable, state of the art deep learning framework that is easy to use, has a flexible programming model, and is well supported by a broad community of academics and industry participants. As a result of recent joint efforts and integration with the latest version of Intel® Math Kernel Library (Intel® MKL), inference performance has increased 100X over the MXNet implementation without MKL. The latest version of MXNet includes built-in support for Intel MKL 2017, which includes support for Intel® Advanced Vector Extensions 2 (Intel® AVX2) and AVX-512 instructions which are supported on Xeon and Xeon Phi processors. All results were presented at AWS re:Invent 2016.

If you’d like to get started with MXNet, instructions for installing and running are available here.

Here are some sample results from running inference benchmarks using MXNet on AWS C4.8xlarge instances[1]:

mxnet-benchmarks-on-aws

 

Co-authored by:
Andres Rodriguez, PhD., Deep Learning Solutions Architect at Intel
&
Niv Sundaram, PhD., Deep Learning Engineering Director at Intel

[1] Source: OS: Linux version 3.13.0-86-generic (buildd@lgw01-51) (gcc version 4.8.2 (Ubuntu 4.8.2-19ubuntu1) ) #131-Ubuntu SMP Thu May 12 23:33:13 UTC 2016. MxNet Tip of tree: commit de41c736422d730e7cfad72dd6afc229ce08cf90, Tue Nov 1 11:43:04 2016 +0800. MKL 2017 Gold update 1