The United States has the fastest supercomputer

May 30, 2022 - 15:50
 0  32
The United States has the fastest supercomputer

The world's most powerful supercomputer has been replaced.

The Frontier is the name of the machine. It is housed in the American Oak Ridge National Laboratory (ORNL) and is known as the first true "exascale" supercomputer due to its output of 1,102 exaflops per second (EFlop / s). The Japanese Fugaku dominated the ranks of the most powerful machines for almost two years.

American Cray, a Hewlett-Packard subsidiary, provides the Frontier supercomputer. Surprisingly, the device's processing capacity is entirely dependent on a single brand's solutions - the American AMD.

The system is built around 64-core Epyc (Milan) processors running at 2 GHz and Instinct MI250 graphics accelerators. Frontier can connect up to 8,730,112 cores.

The latest American supercomputer has an energy efficiency of 52.23 gigaflops per watt. The machine's total performance in the HPL (High-Performance Linpack) benchmark was 1.102 EFlop / s.

In the same test, Fugaku provides 0.442 exaflops per second or 442.01 PFlop / s. The Fujitsu A64FX microprocessors power the Japanese supercomputer, which has a total of 7,630,848 cores.

Fugaku has the capability to exceed the one exaflop limit. However, in the HPL test, which is used to assess the capabilities of supercomputers, Frontier, which is designed to suit the demands of the US Department of Energy, achieved this value.

The new ranking king also moved Finland's Lumi, the most powerful European supercomputer, to third position. It was likewise built by Cray and ran on AMD CPUs. Its output, however, is "only" 151.90 PFlop / s.

Let us not forget that the United States intends to launch an even more powerful machine next year. El Capitan will be built on 96-core AMD Epyc (Genoa) processors.

The second ambitious project is Aurora, which will be based on Intel's competitive Xeons Sapphire Rapids and GPU Ponte Vecchio. Aurora also aspired to be the most powerful supercomputer in the United States.

However, due to the delay in the release of Intel's new chips, other machines eventually surpassed this machine.

PositionSystemCoresRmax (PFlop / s)Rpeak (PFlop / s)Consumption (kW)
1Frontier - HPE Cray EX235a, AMD Optimized 3rd Generation EPYC 64C 2GHz, AMD Instinct MI250X, Slingshot-11, HPE DOE / SC / Oak Ridge National Laboratory (United States of America)8 730 1121102.001685,6521.1
2Supercomputer Fugaku - Supercomputer Fugaku, A64FX 48C 2.2GHz, Tofu interconnect D, Fujitsu RIKEN Center for Computational Science (Japan)7 630 848442.01537.2129,899 th most common
3LUMI - HPE Cray EX235a, AMD Optimized 3rd Generation EPYC 64C 2GHz, AMD Instinct MI250X, Slingshot-11, HPE EuroHPC / CSC (Finland)1 110 144151.90214.352,942 th most common
4Summit - IBM Power System AC922, IBM POWER9 22C 3.07GHz, NVIDIA Volta GV100, Dual-rail Mellanox EDR Infiniband, IBM DOE / SC / Oak Ridge National Laboratory (United States)2,414,592148.60200.7910,096 th most common
5Sierra - IBM Power System AC922, IBM POWER9 22C 3.1GHz, NVIDIA Volta GV100, Dual-rail Mellanox EDR Infiniband, IBM / NVIDIA / Mellanox DOE / NNSA / LLNL (United States)1 572 48094.64125.717,438 th most common