World’s fastest supercomputers
There could be at least three supercomputers that can perform over a billion billion (quintillion) calculations per second in the High-Performance Linpack (HPL) benchmark by the end of 2024.
If each person on Earth could do one of these calculations per second, it would still take over four years to do what these computers can in a second.
Supercomputers can be used for a wide variety of purposes, including scientific research and analysis that works to the benefit of humanity.
Specific fields where they have proven particularly useful include physical simulations, quantum mechanics, molecular modelling, cryptographic system development, weather forecasting, and climate research.
The world’s top supercomputers have rapidly grown much more powerful in the past few years, thanks in no small part to substantial hardware advancements from major chip vendors like AMD and Intel.
Where the fastest supercomputer achieved 442 petaflops per second (Pflop/s) in HP 2020, it has climbed to over 1,000 Pflop/s in 2023.
Below are the world’s top five fastest supercomputers, as they appear on the 62nd edition of the Top500 list of most powerful supercomputers as of November 2023.
Frontier in Tennessee, USA — 1.194 Eflop/s
Deployed and operated for the US Department of Energy, Frontier is the world’s first supercomputer with a performance of over one Exaflop (a billion billion) calculations per second.
It is deployed at the Oak Ridge National Laboratory (ORNL) in Tennessee in the USA, occupying a 680m2.
At its launch in May 2022, ORNL said that Frontier would enable researchers to tackle problems of national and global importance deemed impossible to solve as recently as five years ago.
Using 8.7 million GPU and CPU cores, it posted an HPL score of 1.194 Eflop/s in 2023.
Frontier uses Hewlett Packard Enterprise (HPE) Cray EX architecture that combines the power of 9,472 3rd-generation AMD Epic 64C 2GHz processors with 37,88 AMD Radeon Instinct MI250X accelerators and Slingshot-11 interconnect.
The supercomputer also topped the Green500 list for being the most efficient, with a power consumption of 62.68 gigaflops per watt.
However, it still consumes a hefty 22.7MW of power.
For perspective, Mosselbay’s average electricity consumption is around 20MW.
Frontier Supercomputer. Credit: OLCF at ORNL
Aurora in Illinois, USA — 585 Pflop/s
The world’s second-fastest supercomputer — Aurora — is also operated for the US Department of Energy at the Argonne Leadership Computing Facility (ALCF).
Using only half of the system, the computer achieved a 585 petaflops per second (Pflop/s) score in 2023.
The machine has actually not been completed and is expected to achieve a score of 2 Eflop/s in 2024, making it the new fastest supercomputer in the world.
It is expected to consume 60MW of power when it reaches its intended performance.
The Aurora system is also based on HPE Cray EX architecture but uses 21,248 Intel Xeon Max Sapphire Rapids CPUs coupled with 63,744 Intel Max Ponte Vecchio compute GPUs.
It also features 1.36 petabytes (PB) of on-package HBM2E memory on the CPUs, 8.16 petabytes of HBM2E memory on the GPUs, and 19.9 petabytes of DDR5 RAM.
The supercomputer occupies the same space as two basketball courts — about 840m2.
The plan is to use Aurora for a wide range of workloads, including nuclear fusion simulation, weather prediction, aerodynamics, and medical research.
Aurora Supercomputer
Eagle in Microsoft Azure Cloud — 561 Pflop/s
The most powerful cloud-facing supercomputer in the world, Microsoft’s Eagle is powered by Intel Exon Platinum 84080C 48-core CPUs paired with 14,400 Nvidia H100 GPUs.
It boasts a total core count of about 1,123,200.
Unlike other supercomputers, which require highly privileged access, Microsoft has opened up Eagle usage on its Azure cloud platform for businesses or developers who need to run high-performance workloads.
It is unknown where Microsoft hosts the Eagle supercomputer’s components. As it is cloud-based, the system is possibly distributed across different data centres.
Part of one of Eagle’s nodes. Photo credit: Glenn K. Lockwood
Fugaku in Kobe, Japan —442 Pflop/s
The Riken Centre for Computational Science in Japan hosts this 7.63-million core supercomputer with its highest HPL benchmark score achieved in 2020. It was the world’s most powerful supercomputer for about two years.
Fugaku initially featured 158,976 Fujitsu A64FX microprocessors based on Arm version 8.2A processor architecture. The CPUs were later increased by an unspecified number.
It currently boasts a combined 7.63 million cores.
While there have been reports that China is secretly operating at least one supercomputer with twice the performance of Fugaku, this remains to be confirmed. As it stands, Fugaku is Asia’s fastest supercomputer.
LUMI in Kajaani, Finland — 380 Pflop/s
Europe’s fastest supercomputer uses the HPE Cray EX system and is described as having performance equivalent to 1.5 million of the latest laptop computers.
LUMI stands for Large Unified Modern Infrastructure and was established by the European High-Performance Computing Joint Undertaking (EuroHPC) and a consortium of European countries.
The supercomputer consists of numerous 64-core AMD Epyc CPUs and 10,240 AMD Radeon Instinct MI250X GPUs. It also boasts 1.75PB of memory and 117PB of storage.
LUMI occupies roughly 150m2 of floor space.
Two of its standout features are that it is 100% powered by hydro energy and that its generated waste heat is used to warm up hundreds of homes in the city of Kajaani.
Following recent upgrades, Lumi achieved an HPL score of 380 Pflop/s.
Upcoming: Jupiter — Another Exascale supercomputer
The first exascale computer in Europe is set to come online at the Jülich Supercomputing Centre in Germany by the end of 2024.
Like LUMI, it is being developed by EuroHPC and will support the development of high-precision models of complex systems and artificial intelligence (AI) applications in science and industry.
“Applications will include training large language models in AI, simulations for developing functional materials, creating digital twins of the human heart or brain for medical purposes, and high-resolution simulations of climate that encompass the entire Earth system,” EuroHPC explained.
The supercomputer will use a combination of Arm CPUs and Nvidia GPUs.
Loading ...