The latest TOP500 supercomputing rankings are announced: Frontier retains first place, and Aurora airborne second place
IT Home reported on November 14 that the 62nd TOP500 supercomputer list has now been announced, showing that Frontier still maintains the first position, and is still the only exascale supercomputer in the current public model.
In addition, the top 10 landscape has changed significantly thanks to five new or upgraded supercomputers, with Japan's Fugaku dropping to fourth place, followed by Argonne National Laboratory's Aurora and Eagle installed in Microsoft's Azure cloud.
Frontier, based at Oak Ridge National Laboratory (ORNL) in Tennessee, USA, maintains its lead with an HPL score of 1.194 EFlop/s. Frontier uses AMD EPYC 64C 2GHz processors, based on the latest HPE Cray EX235a architecture. The system has a total of 8,699,904 CPU and GPU cores. In addition, Frontier has an energy efficiency rating of 52.59 GFlops per watt and can be used for data transfer with the help of HPE's Slingshot 11 network.
The Aurora system, located at the Leadership Computing Center in Argonne, Illinois, USA, finished second on the list with an HPL score of 585.34 PFlop/s (previously occupied by Fugaku). However, the Aurora value is submitted in its current state, which is only half of the final size of the plan. It is claimed that the Aurora will surpass the Frontier when construction is completed, with a peak performance of up to 2 EFlop / s.
It's worth mentioning that Aurora is built on HPE Cray EX (Intel Exascale Compute Chip), which uses Intel Xeon CPU Max series processors and Intel data center GPU Max series accelerators, and communicates through HPE's Slingshot-11 network.
There are 20 new supercomputing systems using Intel Sapphire Rapids CPUs throughout the list, bringing the total to 25, making Intel Sapphire Rapids CPUs the most widely used CPU products in new systems. However, only four of the 45 new systems on the list currently use corresponding Intel GPUs, with Aurora being the largest one at the moment.
The third-place Eagle is installed in Microsoft's Azure cloud in the United States, which is also the highest ranking of cloud systems on the TOP500 list. In fact, just two years ago, one of Azure's predecessor systems made it into the top 10, but only ranked 10th. Today, this system, called Microsoft NDv5, has an HPL score of 561.2 PFlops / s and is based on an Intel Xeon Platinum 8480C processor and NVIDIA H100 accelerator.
Fugaku slipped from No. 2 to No. 4, while it had previously held the No. 1 spot between June 2020 and November 2021. The system is located in Kobe, Japan, and has an HPL score of 442.01 PFlop / s. Of course, it's still the highest-ranking system outside the U.S. to publicly open a supercomputer.
The LUMI system at the European High Performance Computing Center (Euro HPC / CSC) in Kajaani, Finland, took fifth place with an HPL score of 379.70 PFlop/s. The system, which is also the largest supercomputing system in Europe, has undergone several upgrades and remains at the top of the list, this time up from the 309.10 PFlop/s HPL score it had last published.
It is worth mentioning that the European High Performance Computing Consortium (EuroHPC JU) is pooling European resources to develop a top-of-the-line exascale supercomputer for processing big data. LUMI is one of the top exascale supercomputers in pan-Europe, located at the CSC data center in Kajaani, Finland.
Overall, China and the U.S. occupy the majority of the overall TOP500 list, with the U.S. leading from 150 systems in the previous list to 161 in this list, while China has dropped from 134 to 104 (note from IT House: the mainland has not submitted a test benchmark for a stronger system for a long time).
In addition, the number one spot in the GREEN500 rankings remains Henri of the Flatiron Institute in New York, USA. The system has an energy efficiency rating of 65.40 GFlops/Watt and an HPL score of 2.88 PFlops/s. Henri is a Lenovo ThinkSystem SR670 with Intel Xeon Platinum and NVIDIA H100 with a total of 8,288 cores, ranking 293rd on the TOP500 list.
Here's a summary of the top 10 TOP500:
Frontier ranked first, the first system in the U.S. to exceed 1 Exaflop/s, located at Oak Ridge National Laboratory (ORNL) in Tennessee, USA, which uses 8,699,904 cores and currently achieves 1.194 Exaflop/s performance. It is based on the HPE Cray EX architecture and combines 3rd Gen AMD EPYC CPU (optimized for HPC and AI), AMD Instinct 250X accelerators, and Slingshot-11 interconnects.
Aurora came in second with an HPL score of 585 Pflop/s. It is installed at the Leading Computing Center in Argonne, Illinois, USA, based on HPE Cray EX and using Intel Xeon CPU Max series processors, Intel data center GPU Max series accelerators, and Slingshot-11 interconnects.
The third-ranked Eagle was installed in the Microsoft Azure cloud, and this Microsoft NDv5 system, based on a Xeon Platinum 8480C processor and NVIDIA H100 accelerator, achieved an HPL score of 561 Pflop/s.
Fugaku ranked fourth, installed at the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan. It has 7,630,848 cores and is capable of achieving an HPL benchmark score of 442 Pflop/s.
The LUMI system is an upgraded HPE Cray EX system installed at the EuroHPC Center at CSC's EuroHPC Center in Finland and is now ranked fifth with a performance of 380 Pflop/s.
The sixth-ranked Leonardo system was installed at a EuroHPC site in CENECA, Italy. It is an Atos BullSequana XH2000 system with a main processor of Xeon Platinum 8358 32C 2.6GHz, an accelerator of NVIDIA A100 SXM4 40 GB, and a four-track NVIDIA HDR100 Infiniband as an interconnect with a Linpack performance of 238.7 Pflop/s.
Built by IBM and based at Oak Ridge National Laboratory (ORNL) in Tennessee, Summit currently ranks seventh in the world in HPL benchmarks, with a performance of 148.8 Pflop/s and 4,356 nodes, each equipped with two POWER9 CPUs (each with 22 cores) and six NVIDIA Tesla V100 GPUs with 80 SMs each, via Mellanox dual-track EDR InfiniBand Stay connected to the network.
The 8th place MareNostrum 5 ACC is a newly installed supercomputing system located at the EuroHPC/Barcelona Supercomputing Center in Spain. This BullSequana XH3000 system uses a Xeon Platinum 8460Y processor with NVIDIA H100 and Infiniband NDR200 to achieve HPL performance of 183.2 Pflop/s.
The ninth-ranked new EOS system is NVIDIA's DGX SuperPOD-based system, based on the NVIDIA DGX H100 with Xeon Platinum 8480C processors, NVIDIA H100 accelerators, and Infiniband NDR400, which achieves 121.4 Pflop/s performance.
The Sierra system, located at Lawrence Livermore National Laboratory in California, is ranked tenth. Its architecture is very similar to that of the seventh-ranked system Summit. It consists of 4,320 nodes, each equipped with two POWER9 CPUs and four NVIDIA Tesla V100 GPUs, achieving 94.6 Pflop/s performance.