One of the features of the 66th HPC User Forum last month was the presentation of the NCSA/Hyperion 2017 industry survey by Bill Kramer, the National Center for Supercomputer Application’s director of the Blue Waters Project and @Scale Programs.
This NSF-funded survey captures “Worldwide Best Practices in Partnerships between HPC Centers and Industrial Users,” in part inspired by the NSF Cyberinfrastructure Division’s interest in supporting NSCI’s tenets of economic competitiveness of companies and communities.
The August 2017 release of the study consisted of in-depth surveys and interviews of 35 leading publicly-funded HPC centers at universities and labs around the world and 31 major industrial organizations that have engaged with those centers.
These findings follow a series of studies and events that reveal modern HPC challenges and successes among industrial users. The first was a 2012 study on the “Impact on Scientific Discovery using Simulation-Based Engineering & Science,” also funded by NSF (Award ID 1013639). While this survey focused on scientific and technical barriers, the 2017 survey addressed cultural and business practices. Neither study looks at public policy. Both surveys were co-developed by Earl Joseph and Steve Conway of Hyperion Research (formerly IDC) and myself.
These studies began in 2010 with a series of International Industrial Supercomputing Workshops, whose founders were Sang Min Lee at KiSTi in South Korea, Michael Resch at HLRS in Stuttgart, Ashok Krishnamurthy at Ohio Supercomputer Center and myself when I was Director of the Private Sector Program at NCSA. This loosely developed community of HPC providers of services to industry represents a significant fraction of the HPC centers that were contacted for this 2017 survey.
The success of the IISW annual workshops influenced a 2015 book that I co-edited with SURFsara’s Anwar Osseyran from The Netherlands (currently Chair of the Council at PRACE, the Partnership for Advanced Computing in Europe). The book explored industrial computing around the world, with 40 contributors from 11 countries.
Case studies include GE, BP, Rolls-Royce, Porsche, Renault and smaller companies, along with HPC center overviews of industrial applications from Finland, France, Germany, Italy, Japan, Belgium, The Netherlands, Korea, Spain, UK and USA. An important chapter on HPC and high-performance data analytics is written by Steve Conway and Chirag Dekate, who were at IDC at that time.
The common theme of these reports is impact. The sub-themes are science, engineering and economics. The business model responsible for this impact is deep vertical engagement with industrial users of advanced computing.
Key takeaways from these surveys and activities are:
- Most large companies are unable to determine whether they need better science or more computing to
achieve greater impact.
- Impact is influenced greatly through deep engagements with teams that understand the science, the
computing AND the business impact.
- HPC centers must address all three aspects if they expect to make a difference beyond basic research.
- Better computing can mitigate and/or focus the need for better science and engineering.
- Most HPC centers do not effectively combine skills in science/engineering, software, HPC and business
engagement in ways that will positively impact industry.
The economic impact I describe is vital in a time of fixed R&D budgets and stagnant government investments. Politely stated: economic impact can ONLY be achieved through efforts that create firm-level ROI. Indeed, macroeconomic country-level gains occur through the efficiencies of the marketplace occupied by companies and for-profit organizations.
Consider one of the most impressive impact statements I have seen. Bloomberg's David Stringer reported in a July 29, 2016 article in The Sydney Morning Herald that “Supercomputers deliver $665 million [US$500] million savings in BHP hunt for oil.” BHP CTO Diane Jurgens stated in the interview that "Crunching exploration data at a centre in Houston was cutting the time needed to produce oil from sites in Trinidad and Tobago to three years from seven." I know more about how this was accomplished than I can divulge here, but please know that supercomputing techniques were used to generate the speed necessary to drive ROI.
So, two surveys, a book and major headlines about industrial ROI occurred in the past few years. These are certainly not the only indicators of industrial innovation through HPC, but they do point to how industry benefits from the proper utilization of HPC and relevant skills.
Community focus on impact fosters innovation. There is no other way.
The2017 NCSA/Hyperion report can be downloaded here:
The Sydney Morning Herald article can be located at:
The Osseyran/Giles book Industrial Applications of High-Performance Computing: Best Global Practices is available at CRC Press (www.crcpress.com Catalogue #K20795) and Amazon (ISBN: 978-1-4665-9680-1).
Merle Giles is currently CEO of Moonshot Research LLC.