Best Practices for HPC Centers and Their Industrial Partners 

What’s the best way for HPC centers in the public sphere to engage with private industry partners to boost the competitiveness of the companies and the larger communities? That question is at the heart of a new study published today by the National Center for Supercomputing Applications (NCSA) in partnership with Hyperion Research. The study was motivated by the trend toward increased public-private collaborations, with both sides – HPC centers and industrial partners – deriving benefits from the relationships, according to Hyperion.

Starting in 2015, Hyperion (then IDC Research’s HPC group) conducted surveys and interviews with 35 publicly-funded HPC centers engaged in supporting industrial partners and 31 major industrial organizations who were working with HPC centers. The study was funded by the National Science Foundation in support of the 2015 Presidential National Strategic Computing Initiative (NSCI) to preserve and strengthen US leadership in HPC. (Editor’s note: The status and future of NSCI under President Trump remains unclear. The executive order has not been canceled, but there were not any NSCI related requests in the FY18 budget.)

Hyperion says the findings affirm what previous studies have shown: “high performance computing (HPC) is a game-changing technology with an ability to accelerate industrial innovation and boost economic competitiveness.” As evidence of this, research conducted by Hyperion on behalf of the DOE indicates that every dollar invested in an HPC project returns $551 in revenue and $52 in profits or cost savings to industrial partner organizations. More on that ROI study here.

The NCSA-Hyperion report, “Worldwide Best Practices in Partnerships between HPC Centers and Industrial Users” sets out the following list of best practices for centers that are serious about supporting industrial partnerships. The authors believe that together these practices constitute an ideal case, although which ones a center should employ will depend on their circumstances.

+ HPC centers that are serious about serving industrial users as well as scientific users should embed this focus into their missions, with appropriate mission-related justification for supporting both sets of users. Doing this sends a strong signal to the center’s staff and to industrial users that industry partnerships are not just a casual or temporary activity. Government agencies that contribute funding for HPC centers’ industrial partnership programs should actively encourage this expanded articulation of the centers’ missions.

+ Choose HPC systems with industry as well as science in mind. Many academic and other HPC centers have no choice but to provide services to their industrial partners on HPC systems that were specified with only scientific users and their workloads in mind. Given the choice, however, centers where industrial partners are an emphasis should procure systems designed to perform well on the full spectrum of the center’s potential scientific and industrial workloads. These centers should and, sometimes do, involve industry in the process of specifying the systems. The same practice applies to software development.

+ Focus on domains the HPC center knows well. It may be tempting, especially at the start of an industry partnership program, to welcome industrial partners from all domains, to show progress. But this study indicates that the odds for success greatly increase when an HPC center limits its industrial partnership program to domains in which the center has strong existing competence.

+ Hire a person with business experience to lead the industrial partnership program. A difficult lesson often learned by centers en route to partnering successfully with industrial users is that placing an experienced, business-oriented person in charge of the program (e.g., a person with industrial consulting experience) can greatly increase the odds of success and decrease the time needed to achieve success. Government and university program managers typically have a much harder time bridging the gap between their worlds and the business world. It also helps to have other staff with industry experience for consulting and support of the industrial projects.

+ Assign experienced people to work with industrial users. Industrial projects typically are mission-critical for the partners and are staffed with some of the companies’ best people. HPC centers should assign some of their own best people to these important projects. The rewards of this peer-to-peer project staffing can be substantial for all parties. It is also important, where possible, to keep the same center staff assigned to the industrial project from start to finish, rather than replacing them in the middle of projects.

+ Involve industrial users in the peer review process. Peer review of industrial applicants for HPC center access should include one or more true peers, i.e., industry representatives who understand the problem in question.

The report also recommends preparing for and addressing cultural gaps, recognizing that nearly half of industrial partners are first-time users of HPC and industrial partners tend to have shorter time frames, harder deadlines and in some cases more stringent requirements for data security. Other recommendations are that centers should should support both open and proprietary research and adopt a streamlined process for intellectual property and contract agreements.

Interestingly, the report points out the potential for confusion over the term industry, noting “some HPC centers and sponsoring agencies limit their definition of ‘industry’ to vendor relationships [but] for a growing majority of HPC centers, the term ‘industry’ refers to the larger universe of industrial (private sector) HPC users.” Hyperion believes that employing the more narrow definition could limit a center’s future funding levels.

Hyperion is forecasting robust growth for the global HPC market: $23.1 billion in 2015 to $30.3 billion in 2021. It further predicts that more than 29 percent of the 2021 total, or nearly $9 billion, will come from industrial users.

“The increasing reliance of industrial partners on HPC has important consequences,” says Hyperion. “One of them, confirmed in this study, is that few large companies that buy HPC systems—much less small and medium-size enterprises (SMEs)—can justify purchasing high-end supercomputers to address their most advanced R&D needs. Hence, the growing use of HPC in the private sector implies not only that more companies will acquire HPC systems, as the study shows, but also that more companies will seek access to high-end supercomputers at publicly supported academic and other HPC centers in their countries.”

Principal Investigator William Kramer of NCSA states: “One of the most interesting observations to me is that the industrial partners and centers both value the HPC expertise that centers provide to the industrial project as much as the actual computing/analysis system access. I am confident that this study will have a very positive impact on the future of HPC center and industrial collaborations.”

The 66 survey interviews were conducted over a 12-month period; it took another six months to analyze the data and produce the 96-page report. Read it in full here.