An exascale computer is a computer that can perform 1018 (about 100) floating-point operations (1 exaFLOPS) per second. Frontier also won the title of most power-efficient supercomputer, performing 62.68 gigaflops per watt.
1. Frontier Supercomputer
Frontier’s performance pushed Fugaku of Japan’s Riken Computer Science Center to second place in two years, Fugaku’s HPL benchmark performance was 442 petaflops. The Top500 said that Fugaku’s theoretical performance is over 1 exaflop, but HPL benchmarks don’t show that speed.
The US Department of Energy’s HPE Cray EX system is equipped with 3rd generation AMD EPYC CPU 873aks 112 cores optimized for HPC and AI. It uses an AMD Instinct 250X accelerator and a Slingshot-11 interconnect.
2. Fugaku
It is a supercomputer built by Fujitsu and has 7,63,848 cores.
3. Rumi
With an HPE Cray EX system, the performance is 151.9 petaflops.
4. Summit
It is an IBM system installed at Oak Ridge National Laboratory in the United States, and its peak performance is 148.8 petaflops. It is composed of 4,356 nodes, and each node is equipped with 22 cores of Power9 CPU and 6 units of Nvidia Tesla V100 GPU. For node connection, Mellanox’s EDR InfiniBand network is used.
5. Sierra
The Sierra at Lawrence Livermore National Laboratory in the US has a similar architecture to the Summit. It consists of a total of 4,320 nodes, and each node is equipped with two Power 9 CPUs and four Tesla V100 GPUs. Performance is 94.6 petaflops.
6th place.
6. Sunway TaihuLight
It is a system developed by China’s NRCPC, and its performance is 93 petaflops.
7. Purlmutter
A system based on the HPE Cray Shasta platform, it is a heterogeneous system that combines an AMD Epic-based node and an Nvidia A100 acceleration node. The performance is 64.6 petaflops.
8. Selence
It is an NVIDIA DGX A100 SuperPOD system based on AMD EPYC processor, NVIDIA A100 and Mellanox HDR InfiniBand. With a performance of 63.4 petaflops, it is installed at Nvidia’s own research facility in the United States.
9. Tenhe-2A
It used an Entel Xeon CPU and a Matrix 2000 DSP accelerator from NUDT of China, and achieved a performance of 61.4 petaflops with about 5 million cores.
10. Adastra
Another HPE Cray EX system. With a performance of 46.1 petaflops, it is the second fastest supercomputer in Europe.
Comments
Post a Comment