June 25, 2022


Technology Forever

Update from the Frontier of Exascale Software program Advancement

Summit supercomputer simulation demonstrates basic principle of Laser Wakefield Electron Acceleration, a laser pulse that types an electron plasma wave

In advance of the scheduled shipment this year of the U.S.’s initially exascale supercomputer, Frontier, at Oak Ridge Nationwide Laboratory, an worldwide staff of software developers led by a University of Delaware professor is doing work on a plasma physics application.

An posting revealed yesterday in the university’s UDaily by Tracey Bryant aspects the work underway by UD’s Sunita Chandrasekaran, assistant professor of computing and details sciences, along with her team, which is just one of 8 operating on purposes for Frontier, an HPE-Cray system run by AMD CPUs and GPUs.

The plasma physics software under progress, PIConGPU (Particle in Cell), is intended to quickly make simulations for next-generation plasma (particle) accelerators, essential to advancing radiation therapies for cancer and to growing the use of X-rays to probe the composition of elements.

“Dr. Chandrasekaran’s PIConGPU crew is an elite group spanning numerous geographic areas, scientific domains and backgrounds,” explained Dr. Nicolas Malaya, technical direct from Superior Micro Products (AMD) for the Exascale Centers of Excellence. “I fully expect this application to make important scientific final results from this workforce in computational science, supercomputing and plasma physics.”

Bryant’s article features a Q&A with Chandrasekaran, below are excerpts:

Q: How is the challenge heading?

Chandrasekaran: Pretty amazing. We are thrilled to have gotten obtain to the new AMD Instinct MI100 (facts centre GPU) accelerator cards. We ran the total PIConGPU on these freshly introduced cards, and in our scientific studies utilizing a one GPU, we observed a 1.4 times increase in pace compared to MI60. This is promising and offers us a whole lot to glance forward to, for the upcoming-generation CPUs and GPUs for Frontier.

AMD Intuition MI100 GPU

The team is applying accelerator playing cards like this from Superior Micro Gadgets (AMD) to velocity the processing of plasma simulations and complete other intensive calculations.

Q: In wanting at these two supercomputing titans, how do you evaluate Frontier’s speed to Summit’s?

Chandrasekaran: Chatting with my collaborator, Dr. Alexander Debus (head of the Centre for Innovative Units Being familiar with [CASUS] at HZDR, a study laboratory based mostly in Germany) assisted me make some observations — simulations like ours with PIConGPU that would choose two months on Summit may finish up taking just one 7 days on Frontier. This also indicates we would now be in a position to run a number of 10-million time-move simulations on Frontier (just about every time step would consider ~50 milliseconds). Time-step simulations enable us to evaluate the procedure of the computer’s electric power program from hour-to-hour intervals, ideal down to thousandths of a second.

Q: Who are your collaborators and what is it like coalescing an intercontinental crew?

Chandrasekaran: My collaborators are from ORNL, HZDR, CASUS, and the Ga Institute of Technology. I have not satisfied 50 % of my workforce in individual, but it feels like we have been functioning alongside one another for decades. We are now a compact household. Please see this webpage for aspects.

Prof. Sunita Chandrasekaran

Once each individual several months, we make confident to examine the team’s, as nicely the project’s, frequent eyesight and objectives to guarantee the short- and very long-time period goals align properly with CAAR (charge assessment and resolution) deliverables. This is specially crucial for an global group like ours. Most of the discussions and conversations are hashed out in excess of email/Slack prior to scheduling a team cell phone contact, specified that there are extra than a few hours of time variation in between the U.S. and Germany.

Q: What is the most interesting/fulfilling element of the venture for you?

Chandrasekaran: I consider it is the interdisciplinary element of this job. It is intriguing to feel about implementing computer system science ideas to a actual-globe scientific software. I am also thrilled that our near collaborations have led to this undertaking becoming funded by Dr. Michael Bussmann (CASUS at HZDR, Germany). This is my to start with internationally funded collaborative undertaking.

Q: What are the areas where Frontier is poised to have the greatest impression? Do you assume Frontier to assist progress upcoming virus analysis, for example?

Chandrasekaran: I think so, particularly when we are in the phase of integrating high-performance computing (HPC), synthetic intelligence (AI) and details science. Large-scale (and speedy) simulations that couldn’t be imagined just a handful of a long time ago are now heading to turn into attainable with the significant compute means that Frontier is going to provide. Not just virus analysis, but such compute abilities are of paramount significance to experiments like obtaining a get rid of for Alzheimer’s disorder or learning weather improve.

Q: How are UD students contributing to the energy?

Chandrasekaran: My Ph.D. university student, Matt Leinhauser, has been operating on this job considering the fact that its inception. With mentorship from myself and my CAAR crew (specifically Rene Widera, Sergei Bastrakov and former CAAR liaison Ronnie Chatterjee), Matt has been able to set jointly two specialized paperwork on profilers — these are applications that recognize parts in the computer system plan that just take the most computation time.  We have so considerably utilized NVIDIA’s nvprof and Nsight profiler applications to dive deeper into the code. HZDR also invited Matt to devote last winter season (January 2020) with them, which was a gratifying chance when he was however in his 1st calendar year of the Ph.D. application.

Q: What is on the horizon?

Chandrasekaran: With guidance from the Frontier Center of Excellence workforce, we will be marching ahead to port PIConGPU on the early accessibility techniques and planning the software for Frontier, which is staying developed as we converse. As future steps, we will be operating on optimizing PIConGPU on the early obtain techniques and rushing up the simulations even further more.