The Top500 ranking of the fastest supercomputers celebrates its 60th edition this November. There is no new leader, but another system from Europe has reached the top 10, along with Leonardo from Italy.
Leonardo enters the top 10
According to the Top 500 ranking list, the US System Frontier remains the world’s fastest supercomputer with maximum computing power in the Linpack benchmark with 1,102 petaflops, or 1.1 exaflops.
The only newcomer to the top 10 is the system leonardo from Italy, which edged out the US Calculator Summit in fourth place with 174.7 petaflops. The Atos Bullsekwana XH2000 supercomputer is equipped with an Ice Lake-SP generation Intel Xeon Platinum 8358, each with 32 cores clocked at 2.6 GHz. However, the lion’s share of computing power is provided by the Nvidia A100 coprocessor in the SXM version with 40 GB of memory (HBM2). 1,374,912 are accounted for by the A100 module, for a total of 1,463,616 “cores” alone.
Leonardo is located at the CINECA data center in Bologna, Italy, and is one of the world’s supercomputers. EuroHPC Project,
In addition to a “booster module” with the above qualities, a “data centric module” is also planned for the Leonardo, which uses Intel’s late Sapphire Rapids processors.
LUMI is now twice as fast
Although still in third place, there were Lumi A big upgrade from Finland. With the AMD Epic Milan and AMD Instinct Mi250X the system size has only been doubled and so it now doubles the computing power to 309.1 petaflops (previously 151.9 pFlops). So the computing power is enough to call itself the fastest supercomputer in Europe.
JUWELS remains Germany’s fastest
The French Adastra. is behind JUWELS BOOSTER MODULE in twelfth place. With AMD Epyc and Nvidia A100 at the Research Center Jülich the system is currently the fastest in Germany and the fourth fastest in Europe with 44.12 petaflops.
But next year work will start on the faster system at the same location. Together with JUPITER, Europe’s first exscale supercomputer is to be built on the premises of the Forschungszentrum Jülich.
Henry leads the Green500 with Nvidia Hopper
Six of the ten most efficient systems use a combination of AMD Epyc Match and AMD Instinct MI250X. But the most efficient system in terms of GigaFLOPS per watt is now Henry from the United States, which combines Intel’s 32-core Xeon Platinum 8362 with Nvidia’s H100-type accelerator card. This results in 65.09 GFLOPS/Watt, a clear edge over the previous leader Frontier TDS with 62.68 GFLOPS/Watt.
ComputerBase received information about this article from Top500.org under NDA. The only specification was the earliest possible publication date.
