What Drives the 2022 TOP500 Supercomputer Rankings? Key Trends and Insights

TOP10 Rankings Overview

Only one new system entered the TOP10: Microsoft Azure's Voyager‑EUS2, ranking 10th with 30.05 PFlop/s. The top spot is still held by Japan's Fugaku (44.2 PFlop/s, 7,630,848 cores) using Fujitsu's custom ARM A64FX. The United States' Summit remains second (148.8 PFlop/s) with dual Power9 CPUs and six NVIDIA V100 GPUs per node.

Other notable entries include:

Sierra (USA) – third, similar architecture to Summit.

Sunway TaihuLight (China) – fourth with 93 PFlop/s.

Perlmutter (USA) – fifth, new entry in TOP10, based on HPE Cray Shasta, AMD EPYC 7763 and NVIDIA A100, achieving 70.9 PFlop/s.

Selene (USA) – sixth, NVIDIA DGX A100 SuperPOD with AMD EPYC and NVIDIA A100.

Milky Way‑2A (China) – seventh, Intel Xeon and custom Matrix‑2000 accelerator.

JUWELS Booster Module (Germany) – eighth, AMD EPYC + NVIDIA A100.

HPC5 (Italy) – ninth, Dell PowerEdge with NVIDIA V100.

Voyager‑EUS2 (Microsoft Azure, USA) – tenth, 48‑core AMD EPYC, 80 GB NVIDIA A100, Mellanox HDR InfiniBand.

Overall System Distribution

China still leads in installed supercomputer count with 173 systems (34.6% of the TOP500), followed by the United States (149) and Japan (32). Over the past two years, China's system count shows a slight decline, while the United States added 27 new systems.

Manufacturer Landscape

Chinese vendors dominate the hardware market: Lenovo (180 systems), Inspur (50), Sugon (36), Huawei (7), together accounting for 273 of the 500 systems. U.S. vendor HPE ranks second with 84 installations. Compared with the previous edition, China's total installations decreased by 21.

Processor and Accelerator Trends

Intel remains the largest processor supplier (81.6% share), though its share fell from 86.4%. AMD's share rose to 14.6%. Among the 70 newly listed systems, 42 use Intel CPUs and 28 use AMD. ARM‑based processors appear in five systems, unchanged from the prior list.

Accelerators are crucial for performance: 151 systems (30.2%) use GPUs or other accelerators. NVIDIA dominates with 84 Volta, 43 Ampere, and 8 Pascal GPUs, totaling 135 NVIDIA‑based systems. Among the 70 new entries, 25 employ NVIDIA accelerators.

Interconnect Technologies

Ethernet is the most common interconnect (48.4% of systems), but InfiniBand provides higher performance for larger machines, present in 178 systems (44.5% overall) and 61 of the top‑100.

Shift Toward Data‑Intensive Computing

Supercomputing is evolving from compute‑centric to data‑centric workloads, driven by AI, machine learning, and big‑data analytics. The convergence of AI, big data, and HPC is reflected in the growing share of "computing‑as‑a‑service" offerings (46% of the TOP100) and the increasing number of systems supporting AI and data‑intensive applications.

Key Takeaways

Fugaku retains the top performance position, Intel and NVIDIA dominate processors and accelerators, InfiniBand remains the preferred interconnect for high‑performance systems, China leads in system count while the United States leads in total performance, and the industry is rapidly embracing AI and big‑data workloads.