World’s quickest supercomputers: Fugaku continues to be No. 1 at 3X the pace of No. 2

The newest semiannual TOP500 record of the world’s quickest supercomputers is topped by Fugaku, the identical machine that gained in June. Constructed by Fujitsu, Fugaku is thrice as quick as its nearest rival.

TOP500 says that competitors for its record appears to be lessening, with the complete record of 500 techniques having the fewest variety of new entries for the reason that group began its monitoring. The record is up to date each June and November and has tracked the event of supercomputer efficiency and structure since 1993. Nonetheless, two model new techniques managed to interrupt into the highest 10 record on their first attempt.

The brand new high 10 record additionally consists of two supercomputers which can be put in in business settings moderately than analysis environments.

The slowest system on the record of 500 scored 1.32PFLOP/sec on the Excessive Efficiency Linpack (HPL) benchmark, up from 1.23PFLOP/sec recorded within the June 2020 rankings. The HPL benchmark measures how effectively techniques remedy a dense system of linear equations.

In an identical vein, the combination efficiency of all 500 techniques grew from 2.22 exaflops in June to 2.43 exaflops on the most recent record.

From a nationwide standpoint, the U.S. continues to dominate the highest 10 with 4 entries.  China has two entries, and Germany, Italy, Japan, and Saudi Arabia every have one. Listed below are the main points.

High500’s 10 quickest supercomputers

The quickest and nonetheless champion is Fugaku, whose high pace bumped as much as 442PFLOP, which is 26PFLOP sooner than in June and thrice sooner than the Quantity 2 system That is due to growing its capability from 7,299,072 cores to 7,630,848 of Arm A64FX processors. The pc was constructed by Fujitsu and resides on the RIKEN Middle for Computational Science (R-CCS) in Kobe, Japan.

At Quantity 2 is Summit, constructed by IBM for the Oak Ridge Nationwide Laboratory (ORNL) in Tennessee with a efficiency of 148.8PFLOP/sec. The system has 4,356 nodes, every with two 22-core Power9 CPUs and 6 NVIDIA Tesla V100 GPUs for a complete of two,414,592 cores. It ranked Quantity 2 in June as effectively.

Sierra retained its Quantity three rating with an HPL rating of 94.6 petaflops. Its structure is just like that of Quantity 2, Summit. Sierra has 4,320 nodes (1,572,480 cores) geared up with two Power9 CPUs and 4 NVIDIA Tesla V100 GPUs.

The Quantity Four slot goes to Sunway TaihuLight. It is powered by Sunway SW26010 processors with 10,649,600 cores and might hit an HPL rating of 93PFLOP/sec. The system was constructed by China’s Nationwide Analysis Middle of Parallel Pc Engineering & Expertise (NRCPC) and is put in on the Nationwide Supercomputing Middle in Wuxi, China. It ranked Quantity 5 in June.

At Quantity 5 is Selene, an Nvidia DGX A100 SuperPOD based mostly on AMD EPYC processors with Nvidia’s new A100 GPUs for acceleration, boosting its efficiency from 27.5PFLOP/sec to 63.4PFLOP/sec on HPL. That moved its rating up from Quantity 7. It’s put in at Nvidia Corp. within the U.S. and has 555,520 cores. (Learn extra: Highflying Nvidia widens its attain into enterprise knowledge facilities)

Tianhe-2A (Milky Method-2A), is ranked Quantity 6 with a efficiency of 61.4PFLOP/sec on HPL. It is powered by Intel Xeon CPUs and NUDT’s Matrix-2000 DSP accelerators and has 4,981,760 cores. Milky Method-2A was developed by China’s Nationwide College of Protection Expertise (NUDT) and is deployed on the Nationwide Supercomputer Middle in Guangzho, China. It ranked Quantity 5 in June.

JUWELS Booster Module makes its debut on the record at Quantity 7. The Atos-built BullSequana machine makes use of AMD EPYC processors with Nvidia A100 GPUs for acceleration, which is analogous to Selene. Put in on the Forschungszentrum Jülich in Germany, it has 449,280 cores and an HPL rating of 44.1PFLOP/sec, making it the quickest in Europe. It’s a part of a modular system, and a second Xeon-based JUWELS Module is listed individually on the TOP500 at place 44. The modules are built-in by the ParTec Modulo Cluster Software program Suite.

Filling the Quantity Eight slot is HPC5, a Dell PowerEdge system with a efficiency of 35.5PFLOP/sec powered by Intel Xeon Gold CPUs and Nvidia Tesla V100 GPUs with a complete of 669,760 cores. It’s the strongest system on the TOP500 record that’s put in at a buyer website and used for business functions. The Italian firm Eni S.p.A. put in it. The system ranked Quantity 6 in June.

Frontera, a Dell C6420 system that was put in final 12 months on the Texas Superior Computing Middle of the College of Texas, is listed at Quantity 9, down from Quantity 8. With its 448,448 Intel Platinum Xeon cores, it hits 23.5PFLOP/sec on HPL.

The Quantity 10 entry, like Quantity 7, is a brand new system referred to as Dammam-7, which is put in at Saudi Aramco in Saudi Arabia. The HPE Cray CS-Storm system is constructed on Intel Gold Xeon CPUs and NVIDIA Tesla V100 GPUs. With 672,520 cores, it reached 22.4PFLOP/sec on the HPL benchmark.