Hpe hub banner.png

USA's 'mini supercomputer' Crusher provides Frontier testbed

Crusher mini supercomputer: Image Credit: ORNL

Image:
Crusher mini supercomputer: Image Credit: ORNL

A miniature version of the USA's upcoming $600 million Frontier supercomputer is now up and running and producing 'highly impressive' results.

Frontier, the United States' first exascale supercomputer, was supposed to become the world's fastest supercomputer when it arrived at Oak Ridge National Laboratory last year, but it is still undergoing integration and testing.

Instead, Crusher is a small but powerful test system, already running four research projects.

In terms of theoretical peak performance, the Frontier system is predicted to provide at least 1.5 exaflops, easily beating Japan's 442-petaflop Fugaku system, which is currently the world's most powerful supercomputer (although China was the first to reach exascale computing).

Frontier was initially due to go live in 2021, but is now scheduled for the second half of 2022, with full user operations beginning in 2023.

Crusher is a 1.5-cabinet version of the 100-plus-cabinet Frontier. It uses 192 HPE Cray server nodes, linked together with HPE's Slingshot Interconnect. Each node uses one optimised 3rd Gen AMD EPYC CPU and four AMD Instinct MI250X GPUs.

Taking up just 44 ft², Crusher is just a fraction of the size of the Department of Energy's 27-petaflop Cray XK7 Titan supercomputer. However, it's faster than the full 4,352 ft² system, which was decommissioned in 2019.

At the moment, four established projects are using Crusher to execute code on Frontier's architecture, with 'outstanding' outcomes. They are: Computational Hydrodynamics on Parallel Architectures (Cholla); CANcer Distributed Learning Environment (CANDLE); Nuclear Coupled-Cluster Oak Ridge (NuCCOR); and Locally Self-Consistent Multiple Scattering (LSMS).

During early testing on Crusher, the NuCCOR team discovered that its computational kernels were up to eight times faster on one of the AMD Instinct MI250X GPUs that power Frontier than one of Summit's NVIDIA V100 GPUs.

The Oak Ridge Leadership Computing Facility (OLCF) said the LSMS team has successfuly run its materials algorithm, which crunches large-scale simulations up to 100,000 atoms, on Crusher and will scale to operate on the whole Frontier system once it is operational.

OLCF director of science Bronson Messer said, "Crusher is the latest in a long line of test and development systems we have deployed for early users of OLCF platforms and is easily the most powerful of these we have ever provided."

He added that the results the teams are achieving are incredibly promising, "[As] we look toward the dawn of the exascale era with Frontier."

In the United States, the Department of Energy's Argonne National Laboratory is also developing another exascale supercomputer, the long-delayed Aurora, which could be even faster than Frontier when it debuts in late 2022.

You may also like

EU to decide fate of $14bn HPE-Juniper Networks merger next month
/news/4333291/eu-decide-fate-usd14bn-hpe-juniper-networks-merger

Mergers

EU to decide fate of $14bn HPE-Juniper Networks merger next month

UK CMA is also investigating the deal

US court clears Mike Lynch of fraud charges
/news/4269317/us-court-clears-mike-lynch-fraud-charges

Law

US court clears Mike Lynch of fraud charges

Lynch, who faced 15 counts of fraud, says he is ‘elated’

Microsoft: Russian hackers are targeting other companies
/news/4167972/microsoft-russian-hackers-targeting-companies

Threats and Risks

Microsoft: Russian hackers are targeting other companies

Nobelium group is linked to Russian intelligence