New IBM Software Toolkit Supercharges Deep Learning

By Charles King, Pund-IT, Inc.  November 16, 2016

A few weeks ago, IBM launched a new POWER-based data center solution for High Performance Computing (HPC) applications, including artificial intelligence (AI), deep learning and advanced analytics. The Power System S822LC is a Linux-based offering that leverages a new POWER8 chip and NVIDIA’s NVLink interconnect technology optimized for the Power architecture. Via NVLink, IBM’s Power server architecture can be tightly integrated with NVIDIA’s Pascal architecture and the company’s Tesla P100 GPUs.

Why is this a big deal? Because the new Power System S822LC solutions avoid the potential bottlenecks that are commonly associated with conventional PCIe interfaces. That’s a good thing in HPC applications that require sustained, muscular data throughput. But it also means that HPC systems utilizing Power System S822LC hardware can deliver considerably higher performance than similarly configured Intel-based systems with PCIe.

IBM and NVIDIA at SC16

Which brings us to the Supercomputing 2016 conference in Salt Lake City this week where IBM announced PowerAI, a new deep learning software toolkit designed to run on one or more Power System S822LC servers. IBM PowerAI supports five deep learning software frameworks, including Caffe by Berkeley Vision and Learning Center (BLVC), one of the popular and most widely used frameworks.

What does this mean in practical terms? According to IBM benchmarks, a Power System S822LC with 4 Tesla GPUs and PowerAI can deliver more than double the performance of comparable x86-based systems running AlexNet with Caffe. The same 4-GPU Power-based configuration running Alexnet with BVLC Caffe can also outperform 8 M40 GPU-based x86 configurations.

IBM says that makes its solution the world’s fastest commercially available enterprise systems platform on two versions of a key deep learning framework, a point that should make the Power System S822LC with PowerAI highly attractive to HPC customers in sectors, including in automotive (for self-driving cars), banking (for facial recognition/fraud detection) and retail (for fully automated call centers).

The impressive scalability of PowerAI—it can run on a single system, as well as in supercomputing clusters with dozens, hundreds or even thousands of servers—should also pique the interest of organizations hoping to leverage deep learning in commercial applications of virtually any size.

Final analysis

The SC 16 announcements come as IBM continues to make considerable progress in cognitive computing and related strategic areas, such as AI, deep learning and advanced analytics by applying leading edge technical innovations to practical business applications and use cases. The sales of Power Systems have obviously benefitted from this approach. In fact, IBM noted strong demand for its Linux-based Power solutions in the third quarter, leading to 2X YoY quarterly growth.

Plus, IBM is serving clients that aim to utilize the company’s solutions to reap significant, even singular benefits. IBM highlighted four customer success stories at SC 16 related to sales of Power System S822LC solutions, including,

  • Human Brain Project, for the group’s new JURON supercomputer
  • Nimbix, for cloud-based services aimed at developers and data scientists
  • The city of Yachay, Ecuador, for the country’s first supercomputing cluster, and
  • SC3 Electronics, for what will become the largest HPC cluster in the Middle East and North Africa

IBM’s strategy of solving acute technical challenges while delivering quantifiable business benefits has enabled it to establish a strong competitive position in cognitive computing during 2016. Absent significant changes in the market or challenges from competitors, IBM should deliver more of the same in 2017.

© 2016 Pund-IT, Inc. All rights reserved.