A new study highlights the importance of locally available supercomputers to university research. Authored by a cross-disciplinary team of experts from Clemson University, the report provides compelling evidence connecting TOP500-level computing power with technical efficiency of research output.
HPC users are well aware of the advantages that leadership computing confers in terms of competitiveness, innovation and other societal benefits, but policy makers require repeated assurances that the large expenditures are justified. The aim of the Clemson study is to put hard numbers behind assumptions of HPC’s merits. To do this, the authors developed a quantitative economic model that compares HPC “haves” and “have-nots” across a range of disciplines.
Per the study design, 212 institutions were classified into “haves” — those with a TOP500 supercomputer — and “have-nots” — those without TOP500-level power.
To evaluate technical efficiency, the study relied on input variables — listed as the total number of faculty members and incoming graduate students’ average GRE scores — and output variables — the total number of publications for the academic year and the number of Ph.D. degrees awarded.
The research team, led by Amy Apon who chairs the Computer Science Division in the School of Computing, found the biggest effect of having a world-class supercomputer in the fields of chemistry, civil engineering, physics and evolutionary biology. Chemistry “haves” were doubly efficient compared with “have-nots,” and in civil engineering, a TOP500 supercomputer provided a 35 percent efficiency edge. Physics and evolutionary biology were also positively impacted but to a lesser degree.
The supercomputing effect did not extend to all domains, however; research output was not enhanced for computer science, economics or English — and for biology, results were mixed, according to the Clemson colleagues.
“For the nation, it is unequivocal that a high-performance computing system will provide an advantage in doing research in several fields,” stated Apon. “It’s not uniform across all fields. But for fields where it matters, it matters a lot.”
Paul W. Wilson, the lead economist on the study, sees it as a potential tool for helping policy makers with investment decisions relating to science and innovation as well as cyber-infrastructure.
“While many would agree that high-performance computing has a positive effect on research output, the connection has been assumed and qualitative until now,” he said. “This is a critical first step in creating a model for evaluating investments in high-performance computing.”
Jim Bottum, Clemson’s chief information officer and vice provost for computing and information technology, is encouraged by the results. Clemson is one of the supercomputing “haves” owing to its Palmetto Cluster, ranked as the sixth fastest public university supercomputer in the US on last November’s TOP500.
“Our research results provide a critical first step in a quantitative economic model for investments in HPC,” the authors wrote in the journal article describing their research, published in a special issue of Empirical Economics.