Frontier Breaks the Exaflop Barrier: Inside the New Top500 Supercomputer Landscape

Frontier Achieves True Exascale Performance

On May 30, 2022, the 59th ISC (International Supercomputing Conference) released the latest Top500 list, marking a new era in high‑performance computing as ORNL's Frontier became the first system to surpass the exaflop threshold with a measured Linpack performance of 1.102 exaflops.

Frontier consists of 74 Cray EX cabinets housing 9,408 nodes. Each node combines an AMD Milan "Trento" 7A53 EPYC CPU with four AMD Instinct MI250X GPUs, totaling 37,632 GPUs. Nodes are interconnected via HPE's Slingshot‑11 network. Every node provides 512 GiB DDR4 memory and 512 GiB HBM2e (128 GiB per GPU) shared across nodes.

Top10 Rankings and New Entrants

The new Top500 list features a diverse set of systems:

Frontier (USA) – 1.102 exaflops (rank 1)

Fugaku (Japan) – 442 petaflops (Arm A64FX, rank 2)

LUMI (Finland) – 151.90 petaflops Linpack, 214.3 petaflops theoretical peak (rank 3)

Summit (USA) – rank 4

Sierra (USA) – rank 5

Sunway TaihuLight (China) – rank 6

Perlmutter (USA, HPE Cray EX) – rank 7

Selene (USA, NVIDIA) – rank 8

Tianhe‑2A (China) – rank 9

Adastra (France) – rank 10, 46.1 petaflops Linpack, 61.6 petaflops theoretical peak

The three newly added systems—Frontier, LUMI, and Adastra—share the HPE Cray EX235a architecture.

Chinese Supercomputing Progress

China currently operates three E‑class supercomputers that are not Top500‑qualified but are recognized by the Gordon Bell Prize. These include the existing Sunway TaihuLight (now located in Qingdao) and Tianhe‑3, which uses Phytium 2000+ FTP Arm CPUs and Matrix 2000+ MTP accelerators, delivering an estimated 1.7 exaflops peak and over 1.3 exaflops Linpack performance.

According to insiders, China plans a 10‑exaflop system for 2025‑2026, likely based on the Sunway architecture (Alpha‑core derived).

Technology Suppliers and Market Share

Lenovo appears most frequently on the Top500, contributing 17 new entries. HPE follows with 14 new systems, seven of which use AMD CPUs and GPUs connected via Slingshot‑11. Overall system counts are Lenovo (180), HPE (84), and Inspur (50). In terms of performance share, HPE leads with 18.6 %, Fujitsu follows with 18.1 %, and Lenovo holds 15.1 %.

NVIDIA is present in 19 systems and collaborates on five additional ones, including Sierra and Selene. Intel’s share has dropped to 77.40 % from 81.60 % six months earlier, while AMD’s share rose to 18.80 % from 14.60 %.

IBM maintains nine systems, including Summit and Sierra, and continues to partner with Lenovo on the SuperMUC Phase 2 system.

Key Takeaways

Frontier’s breakthrough dramatically widens the performance gap at the top of the Top500 list, while the overall distribution shows strong representation from the United States (9 new systems) and Germany (5 new systems). China, despite not adding new entries this cycle, still operates the largest absolute number of systems (173) worldwide.

These trends underscore the accelerating race toward exascale computing, the growing importance of heterogeneous architectures (CPU + GPU), and the strategic role of major vendors such as HPE, Lenovo, and NVIDIA in shaping the future HPC landscape.