Share
FollowNERSC Hosts First ‘GPUs for Science’ Workshop
July 12, 2019
Day 1 featured more than 20 presentations on programming models, performance portability, data analytics, and a variety of application/use cases, including the superfacility model, bioinformatics, and machine learning.
More than 300 people from multiple national laboratories, academia, and industry attended the inaugural “GPUs for Science” workshop at Berkeley Lab, held July 2-3, 2019.
When Perlmutter, the National Energy Research Scientific Computing Center’s (NERSC) next supercomputer, arrives in 2020, researchers computing at NERSC will need to be prepared to use GPUs for their simulation, data processing, and machine learning workloads. Toward this end, this event – which was the idea of, and conducted by, a group of NERSC postdoctoral fellows – was designed to facilitate the transition to GPU systems by giving attendees the motivation, tools, and expertise needed to make this change.
“This event was really the brainchild of the NERSC postdocs,” said Jack Deslippe, acting lead for NERSC’s Application Performance Group. “A lot of the excitement around GPUs for science is coming from some of the staff who are newest at NERSC. They are really driving this.”
Day 1 featured more than 20 presentations on programming models, performance portability, data analytics, and a variety of application/use cases, including the superfacility model, bioinformatics, and machine learning.
“The general feedback we got was that the workshop was very helpful,” said Yunsong Wang, a postdoc in NERSC’s NESAP for Data group who helped organize the event, along with fellow NESAP postdocs Laurie Stephey and Rahul Gayatri and former NESAP postdoc Jonathan Madsen, who is now an application performance specialist at NERSC. “Most of the attendees were domain scientists new to GPUs, so the Day 1 presentations were really useful to them.”
Day 2 continued with lightning talks on Python, ptychography, AMReX, WarpX, and BerkeleyGW; several tutorials, including a hands-on tutorial by NVIDIA engineer Max Katz on GPU profiling; and a hacking competition using the GPUs on NERSC’s Cori system.
"At present, there are 18 nodes integrated into Cori that each have 8 GPUs, so we have a total of 144 GPUs,” Deslippe said. “This is a resource that NERSC is using to work with application teams in preparation for the upcoming Perlmutter system."
Toward this end, NERSC plans to hold a number of GPU training events, hackathons, and competitions over the next few years, he added. “One of the big motivations is to begin to energize our community around GPU science and the upcoming system.”
Indeed, energy was abundant throughout the recent workshop. “It was exciting to see the enthusiasm of the attendees,” Stephey said. “Hackathon participants stayed late and were competing right up to the very end to win the contest.”
Slides for many of the presentations from the workshop can be found on the GPUs for Science web page.
About NERSC and Berkeley Lab
The National Energy Research Scientific Computing Center (NERSC) is a U.S. Department of Energy Office of Science User Facility that serves as the primary high performance computing center for scientific research sponsored by the Office of Science. Located at Lawrence Berkeley National Laboratory, NERSC serves almost 10,000 scientists at national laboratories and universities researching a wide range of problems in climate, fusion energy, materials science, physics, chemistry, computational biology, and other disciplines. Berkeley Lab is a DOE national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California for the U.S. Department of Energy. »Learn more about computing sciences at Berkeley Lab.
