Blurring the Lines Between HPC and AI @ SC18

By Gabor Samu, Product Marketing, IBM

December 11, 2018

The dominant topic at SC18 was the convergence of HPC and Artificial Intelligence (AI) with some of the biggest research and enterprise HPC users providing perspectives on how HPC and AI are moving closer together.

One perspective came from Fred Streitz, Chief Computational Scientist and Director of the HPC Innovation Center at Lawrence Livermore National Laboratory.  During a conversation with HPCwire he noted “It’s interesting to see more and more people talking about this convergence of AI and HPC. At Lawrence Livermore, we’re using machine learning to actually guide the decisions that are made about what simulation to run next…What’s the next best simulation to run, given everything you know? And then, once you run that simulation, what’s the next best simulation to run, given everything that you NOW know. We’re doing this in a closed loop on Sierra. It’s a capability that we created in partnership with the National Cancer Institute and with the researchers at IBM. “

In this HPCwire video, Mr. Streitz discusses this new approach to HPC simulation while standing near an IBM booth version of the Summit supercomputer now operational at Oak Ridge National Laboratory (ORNL). Over the past few years, IBM, NVIDIA, Mellanox, and Red Hat all worked together to create Summit at ORNL and also the Sierra supercomputer that Mr. Streitz works with at Lawrence Livermore. Today, Summit is the most powerful and smartest supercomputer for science in the world, and Sierra is a close second. These state-of-the-art supercomputers were not developed as highly customized, one-off systems; instead, they were assembled using every day off-the-shelf components leveraging an architecture that delivers massive HPC scalability purpose-built for AI.

Researchers are using Summit and Sierra to simulate quantum mechanics materials, explore evolutionary relationships based upon genomes, and fabricate nanomaterial using AI, among many other jobs. “By building these supercomputers, we are building the world’s leading AI machines,” says Hillery Hunter, IBM Fellow and Director of Accelerated Cognitive Infrastructure at IBM.

The HPCwire video crew also found Dr. Rosemary Francis at the IBM booth. Dr. Francis is the CEO of Ellexus, a data input/output (IO) profiling company and featured IBM business partner. She explained:  “We make tools to help people understand how they access data. Our customers use HPC in many different ways.”

Concerning HPC – and AI in particular – Dr. Francis noted: “In terms of AI, I think it’s a really interesting challenge for HPC. It comes with very different IO patterns and very different compute patterns. It’s going to shape the availability of hardware and software systems. AI workloads bring a lot in terms of scale – the scale you need for AI systems is going to be a really interesting challenge for HPC.”

Though the Summit supercomputer was a major attraction, and the convergence of HPC and AI the leading topic of discussion, there were dozens of other displays and activities happening in and around the IBM booth at SC18.

IBM has been working on a new strategy called HPC Transformation that incorporates Artificial Intelligence (AI) methodologies into classic approaches to HPC. The IBM theme for SC18 – Blurring the lines between HPC and AI – reflects this new HPC strategy.

For example, at SC18 IBM introduced what is essentially a mini-version of Summit and Sierra called the IBM Power Systems Accelerated Computing Platform (IBM Power ACP). Thanks to the fact that these supercomputers were built using commercially available components, IBM ACP solutions are now available to organizations that want the computational power of Summit or Sierra but don’t have the budget and resources of a national lab.

Many of the individual elements of the Summit/Sierra supercomputers and the new IBM ACP solutions were on display as well, including IBM Power Systems AC922 servers (the building block of Summit/Sierra), IBM Elastic Storage Server (the data storage systems used), and
IBM Spectrum Computing family members such as IBM Spectrum LSF deployed in HPC environments around the world and also leveraged in Summit and Sierra.

2018 was the 30th anniversary of the Supercomputing Conference. During that time, supercomputing itself, as well as the technology and solutions provided by IBM, have changed dramatically. 30 years ago, Deep Blue was winning chess matches against world champions. Ten years ago, IBM supercomputers were running neural simulations equivalent to a cat’s brain. Today, HPC is leveraging the new power of AI to tackle challenges ranging from global climate change to sub-atomic quantum physical effects.

And the icing on the cake?  IBM received a number of HPCWire Awards during SC18 including:

  • HPCWire Readers’ Choice Awards for Top Supercomputing Achievement
  • HPCWire Editors’ Choice Awards for Top Supercomputing Achievement & Top 5 Vendors to Watch

And for an encore?  Well, you’ll just have to attend SC19 and visit the IBM booth to find out!

 

Return to Solution Channel Homepage
HPCwire