Petascale Cloud Computing

Cloud image

A recent study shows that EDGE can sustain petascale performance in the public cloud. By using a total of 27,648 c5.18xlarge cores, the solver reached a 32-bit non-zero performance of 1.09 PFLOPS when simulating seismic wave propagation. This corresponds to a hardware utilization of over 20% at an assumed stable AVX512 frequency of 2.9 GHz. Such performance was previously only possible through on-premises bare-metal supercomputing infrastructure. We will present details on high performance cloud computing at ISC High Performance 2019 with EDGE as application. Stay tuned for the date and time of the technical paper presentation. Documentation on the AWS and GCP cloud setups is available from EDGE’s user guide.

Visualization of fused simulations in an SGT-setting. Visualization of fused simulations in an SGT-setting. The gray spheres indicate the locations of the surface point sources. Colors denote the South-North particle velocities of eight South-North point forces after eight simulated seconds. Warm colors denote positive velocities, cold colors negative ones. The entire run covered 16 fused settings. A recent study shows that public cloud solutions can run similar settings at high performance.

Supercomputing Conference (SC18)

SC18 logo

EDGE is part of the poster presentation Tensor-Optimized Hardware Accelerates Fused Discontinuous Galerkin Simulations at this year’s Supercomputing Conference (SC18) in Dallas, Texas. The poster reception is on Tuesday (11/13/18) from 5:15PM - 7:00PM. Electronic versions of the poster and extended abstract are available from EDGE’s resources.

SC18 poster abstract on EDGE

2018 SCEC Annual Meeting

SCEC logo

September is the month of the SCEC Annual Meeting. Location of the meeting is the middle of the desert: Palm Springs, CA. The meeting is preceded by a day of workshops, where the SCEC Open-Source Software and Data Access Workshop is of particular interest to modeling and simulation. EDGE is part of the poster “Fused Earthquake Simulations on Deep Learning Hardware” (#289), available electronically from http://short.dial3343.org/scec18pos.


All Posts