February 25, 2021
Staff Accomplishment

Krishnamoorthy’s Team Wins Best Student Paper at SC20

The paper provides a new approach for solving a long-standing problem in high-performance computing

PNNL Computer Scientist Krishnamoorthy

PNNL computer scientist and laboratory fellow Sriram Krishnamoorthy's student-led research project won Best Student Paper Award at the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC20).

(Photo by Andrea Starr | Pacific Northwest National Laboratory)

Third time’s a charm for Physical and Computational Sciences Directorate computer scientist and laboratory fellow Sriram Krishnamoorthy, whose student-led research project won Best Student Paper Award at the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC20), a premier annual conference for (HPC) organized by IEEE.

In 2014, Krishnamoorthy was a finalist for both Best Paper and Best Student Paper, but last year was the first time his team received the award.

University of Utah students Arnab Das and Ian Briggs co-authored “Scalable yet Rigorous Floating-Point Error Analysis” in collaboration with their advisor Ganesh Gopalakrishnan and mentors Krishnamoorthy and University of Utah professor, Pavel Panchekha.

“It’s pretty exciting to be part of a winning team, it’s quite an honor to get recognized at SC20 but it’s especially good for the students who put in the effort to make this happen,” said Krishnamoorthy.

Winning in any category at a prestigious conference like SC20 is an indicator of excellence in research, but supporting and mentoring the winners of Best Student Paper specifically highlights Pacific Northwest National Laboratory’s commitment to supporting the next generation of innovators in computer science.

“An award like this brings visibility and recognition from the scientific community that your ideas and work are contributing to the field; it’s really the icing on the cake,” said Krishnamoorthy, who also holds a joint appointment in the School of Electrical Engineering and Computer Science at Washington State University.

The paper provides a new approach for solving a long-standing problem in HPC: detecting errors in the finite precision representation of real numbers produced by computers in a way that is flexible and cost-effective, without losing rigor. The research team presented a tool called SATIRE that scales error analysis by four orders of magnitude compared to today’s best-of-class tools.

“This analysis really hits a sweet spot; we eliminate some of the hand-wringing and maximize the precision in the process,” said Krishnamoorthy.

Published: February 25, 2021