Frontier, the HPE Cray exa-scale supercomputer operated by the Department of Energy (DoE) at Oak Ridge National Laboratory (ORNL), remains by far the fastest supercomputer on the planet, according to the Top 500.
With a High-Performance Linpack (HPL) benchmark score of 1.102 Exaflops per second, Frontier is the only exa-scale supercomputer and is three times faster than its nearest competitor, Fugaku, at the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan, according to Top 500. One exaflop is a quintillion calculations or floating point operations (flops) per second.
Frontier, a HPE Cray EX system, is quipped with 8,730,112 cores and combines 3rd Gen AMD EPYC CPUs optimized for HPC and AI, with AMD Instinct 250X accelerators, and Slingshot-10 interconnect.
Frontier earned top place in Top 500 June ranking, when it achieved the same 1.1 exaflops of performance. It features a theoretical peak 2 exaflops and is 10 times more powerful than ORNL’s Summit system.
Also: The White House passes an AI ‘Bill of Rights’ that attempts to put your concerns at ease
Frontier is also part of the DoE’s Exacscale Computing Project. It’s housed within 74 giant HPE Cray EX supercomputer cabinets, which includes more than 9,400 AMD-powered nodes and 90 miles of networking cables, according to ORNL. It also features 37,000 GPUs.
As ORNL explains, modelling cancer cells, supernovas, the coronavirus or the atomic structure of elements require 64-bit precision afforded by typical supercomputers, whereas machine-learning algorithms require less precession, for example 32-, 24- or 16-bit accuracy and thus can take advantage of GPUs.
Frontier’s Orion storage system holds 700 Petabytes of data; each cabinet weighs 8,000 pounds and the system is cooled by 6,000 gallons of water pumped each minute by four 350-horsepower pumps. The pumps could fill an Olympic-sized pool in 30 minutes. “Frontier is the clear winner of the race to exascale, and it will require a lot of work and innovation to knock it from the top spot,” said the Top 500 press release.
The fastest supercomputer in Europe, in third place in the Top 500, is the Finland-based upgraded LUMI system, also an HPE Cray EX system. It’s installed at the EuroHPC center at CSC’s data centre in Kajaani, Finland and has a performance of 309.1 Pflop/s.
As Top 500 notes, the European High-Performance Computing Joint Undertaking (EuroHPC JU) is pooling European resources to develop top-of-the-range Exascale supercomputers for processing big data.
A new entrant is Leonardo, installed at the EuroHPC site at Cineca, in Italy. It’s a French-made Atos BullSequana XH2000 system with Xeon Platinum 8358 32C 2.6GHz as main processors, NVIDIA A100 SXM4 40 GB as accelerators, and Quad-rail NVIDIA HDR100 Infiniband as interconnect. It achieved a Linpack performance of 174.7 Pflop/s.
Atos earlier this year unveiled the XH300, a more powerful system than XH200 with CPUs, GPUs and accelerators from AMD, Intel and Nvidia, as well as microprocessor tech from French firm The Silicon Pearl (SiPearl).
Old Summit, an IBM-built system, is in fifth spot with a performance of 148.8 Pflop/s.
Sierra, the sister of Summit at the Lawrence Livermore National Laboratory, California, is in sixth place with a score of 94.6 Pflop/s.
China’s fastest supercomputer, Sunway TaihuLight, was the world’s top performing supercomputer in 2017, but now ranks seventh with a performance of 93 Pflops/s. Tianhe-2A (Milky Way-2A) was the second fastest in 2017 but is now in 10th place with a performance of 61.4 Pflop/s. They were eclipsed by Summit and Sierra in 2018.
While China’s fastest supercomputers have slipped in the Top 500 HPL benchmark rankings, China still dominates the Top 500 list along with the US. The US has 126 machines in the Top 500, while China has 162 in this ranking. However, China’s systems count fell from 173 in the June 2022 ranking. Europe as a whole accounted for 131 machines on the Top 500, up from 118 machines in June.