Advanced Computing in the Age of AI | Thursday, March 28, 2024

Industry Voice Added to U.S. Drive to Exascale 

A federally-funded effort to deliver HPC systems 50 times faster than today's supercomputers has added a business perspective to the multi-year development project, which one industry observer said could reduce the runtime for complex simulations from a year to two hours and "really ramp-up U.S. industrial R&D.”

The Exascale Computing Project (ECP) has formed an Industry Council comprised of executives from major U.S. corporations across a range of industries and chaired by an SVP at United Technologies. ECP is led by six Department of Energy national labs, with project management at Oak Ridge National Laboratory.

According to Steve Conway, Research VP, HPC, International Data Corp., the formation of the council highlights the value of supercomputing to the R&D work of American corporations, something that is often under-appreciated.

“A lot of people, even some senior government officials, think that getting to exascale computing is only about science, only about basic scientific research – which is very important – but it’s also about helping America's industrial and economic competitiveness,” Conway told EnterpriseTech. “Some of America’s biggest companies…really depend on HPC for their advanced research that is incredibly important to their ability to compete with companies outside of the U.S. But even small- and medium-sized businesses are going to benefit from exascale… The technical advances that will be happening inside this program in order to reach exascale are going to benefit companies that only use one or two racks of HPC systems, and also companies that just use the cloud to do their HPC… I don’t think that’s very well known."

Conway said calculations that might take a year to process using today’s supercomputers could potentially be run in two hours or less on an exascale system. This means highly complex simulations related to cancer research and aircraft design, to name two examples, could be run at much higher levels of resolution than is possible today. “This is in contrast for people who assume that what these big computers are all about is figuring out whether a proton turns left or right,” he said.

The companies participating in the ECP Industry Council include:

  • Altair Engineering, Incorporated
  • ANSYS, Incorporated
  • Cascade Technologies, Incorporated
  • Chevron Energy Technology Company
  • Cummins Inc.
  • The Dow Chemical Company
  • DreamWorks Animation
  • Eli Lilly and Company
  • Exxon Mobil Corporation
  • FedEx Corporation
  • General Electric
  • General Motors Company
  • Mars, Inc.
  • Procter & Gamble Company
  • Tri Alpha Energy, Inc.
  • United Technologies Corporation
  • Westinghouse Electric Co.
  • Whirlpool Corporation

According to a prepared statement from ECP, the council “will provide guidance and feedback on ECP’s strategic direction, project scope, technical requirements and progress, providing the perspective of private industry as it relates to the emerging need for exascale-level computation and the formation of a holistic exascale ecosystem.”

Dr. J. Michael McQuade, Senior Vice President, Science & Technology, United Technologies Corporation, will serve as the first chair of the council. In discussion the mission of the council, he emphasized ECP’s mission to create an “ecosystem of expertise” around exascale computing, along with its important to American industries competitiveness.

“The big change the way this program is being approached is that the team is also looking at the entire ecosystem,” McQuade told EnterpriseTech, “so it’s not just counting FLOPS and bytes, it’s also looking at the throughput of the machine, what’s the use of the machine, what’s the ability to get code to run through the machine.

“Not only do you have to have fast machines,” he said, “you have to have people who know how to use them, you have to have the right tools for using them, and you have to have a community that interacts to use those machines as efficiently as possible. So one of the things I like about the way this project is structured is that it’s recreating a very large environment of people who represent the type of expertise we’re going to need in the future in this country.”

He said UTC has long been a heavy user of HPC for the design and production of machinery used in the aerospace and other heavy industry markets. The company has relied for years on supercomputers made available to industrial companies by the national labs, including Oak Ridge National Laboratory’s Titan system.

Asked how he’d feel if competing companies in other countries had earlier access to an exascale system, McQuade said, “I’d be a very unhappy guy. It’s a global world, we compete globally. We all compete based on the quality of the tools and the capabilities we have available to us, and if we’re not at the leading edge, as we have been at the leading edge for a couple of decades, we’re going to be at a competitive disadvantage.”

According to ECP Director Paul Messina, “The external Industry Council is vitally important to keeping the project in sync with the real-world needs of the HPC industrial user community. These experienced executives will bring deep insight to the requirements of the U.S. industrial sector and help us ensure future exascale capabilities are designed to address a wide range of industrial applications.”

Conway pointed out that even for the largest industrial companies, it’s not economically feasible to buy and maintain a world-class supercomputer.

“A lot of those companies listed in the announcement use DoE supercomputers for their most advanced research because no matter how big these companies are, they can’t justify buying a gigantic supercomputer for the part of it that they would use,” Conway said. “So this is a great boon to the big US industrial firms, and soon that’s going to include companies that are involved in healthcare and lots of other endeavors that depend on very big data.”

EnterpriseAI