National Energy Research Scientific Computing Center (NERSC) GPUs for Science Day, October 25, 2022, 9 Nov 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Introduction and Welcome

Description

Jack Deslippe (NERSC)
Introduction and Welcome

A

Yeah so so hi everybody I'm, uh I'm, Jack, deslip I'm, the application performance lead at nurse, and it's my uh real pleasure here to to welcome you today to the the gpus for science day.

A

This event uh started about four years ago as an idea at a lunch with myself, uh the nurse division director and our division postdocs- and you know, I- think the post-ops are really excited about the upcoming prom letter system at the time and the potential of nurse first GPU accelerated system to impact science, and so that group of postdocs proposed this idea for GPU for science day.

A

They they planned it and organized the event, assembling I think a really fantastic set of speakers um to cover a whole bunch of different topics in um you know GPU science acceleration, so nurse has held two such GPU for science, Day events in this way and uh after a two-year break, we are back with the third installment and um you know, I think it's never been more clear that gpus really dominate the HPC landscape.

A

um There's now systems at um nurse and Oak, Ridge and Argonne leadership. Computing facilities are all Fielding systems powered by GPU accelerators.

A

um You know from from all of Nvidia AMD and Intel, including the number one system in the world now from Frontier, um and so we are here today to both celebrate the science that has uh kind of made good on the promise of gpus and scientific research and Discovery, and also to prepare to make the most of the opportunities yet to come.

A

So the majority of the agenda will showcase about 10 science, application teams, um sharing their first-hand experiences and best practices on porting to gpus, and we hope that this knowledge and Lessons Learned can be really adapted to users. Own applications um and uh we'll hear today from three major compiler teams working to harness gpus, we'll hear about multiple programming languages and Frameworks for programming gpus, including performance, portable techniques to Target, multiple GPU architectures.

A

um One thing I want to highlight is that the space of GPU Technologies is now pretty vast, so I'll I'll kind of apologize ahead of time for all of the worthy topics in the space that we will not cover today.

A

This will sort of be but a taste of the thriving HPC GPU ecosystem that that has emerged, um and so I want to thank the sort of the the new generation of early career staff at nurse who have led the organization of today's program, so that includes Neil meta, juva, Cole Carney so have gauche, Felix, Whitmore and, of course, I want to thank Helen.

A

He, our training, lead here at nurse for organizing this event and really a week full of great events uh this week and I would encourage you all to return tomorrow again for uh the the the first of two days of day to days, um and so since our last GPU for science day, the pro Mudder system at nurse has um uh been deployed and our user Community has really quickly taken to the system in a big way.

A

We've seen uh Gordon Bell finalist projects utilize the system we've seen high impact science results such as the first EXA op, app initial molecular Dynamics calculations and we've seen super facility data analysis acceleration.

A

uh We see transformative AI methods, utilizing the machine, learning capabilities of gpus in many different science areas, um and you know we designed Pro Mudder as a system optimized for Science, and so it's really exciting.

A

uh You know tremendously exciting I think to see to see that play out um so again welcome to GPU for science day and I. Think without further Ado I'd like to pass the Baton over to J srinivasan. Who is the promoter Project Lead? To actually tell you a little bit about the Pearl Mudder, the Pearl meter system itself, so I think uh take it away. Jay.