 |
 |
 |
Award Finalist/Winner |
SC Conference - Activity Details
0.374 Pflop/s Trillion-particle Particle-in-cell Modeling of Laser Plasma Interactions on Roadrunner
Authors:
|
Kevin J. Bowers
(Los Alamos National Laboratory)
|
|
Brian J. Albright
(Los Alamos National Laboratory)
|
|
Benjamin K. Bergen
(Los Alamos National Laboratory)
|
|
Lin Yin
(Los Alamos National Laboratory)
|
|
Kevin J. Barker
(Los Alamos National Laboratory)
|
|
Darren J. Kerbyson
(Los Alamos National Laboratory)
|
ACM Gordon Bell Finalists Session
|
Thursday, 10:30AM - 11:00AM
|
|
Room Ballroom G
|
Abstract:
We demonstrate the performance and scalability of VPIC on the heterogeneous IBM Roadrunner supercomputer at Los Alamos National Laboratory. VPIC is a 3d, relativistic, electromagnetic, particle-in-cell (PIC) code that self-consistently evolves a kinetic plasma. We conducted VPIC simulations of laser plasma interaction at unprecedented fidelity and scale---over $10^{12}$ particles and $10^8$ voxels---to model phenomena in inertial confinement fusion experiments. For studies of laser reflectivity versus intensity under experimentally realizable conditions, we measured sustained performance exceeding $0.374$~Pflop/s (sp), the inner loop achieved $0.488$~Pflop/s (sp), enabling first-principles modeling of an issue critical to the DOE/NNSA National Ignition Facility. Notably, given data motion limitations, we achieved this with PIC---a technique that typically requires more data motion per computation than techniques often used to demonstrate supercomputers (like dense matrix, molecular dynamics and Monte-Carlo calculations).
|
|
|