HPE dominates TOP500 with trio of exascale leaders

Hewlett Packard Enterprise has built the world's three fastest exascale supercomputers and six of the ten fastest verified systems overall, according to the latest TOP500 rankings. The group also built half of the 20 most energy-efficient machines on the Green500 list.

The newly published rankings confirm that El Capitan, installed at Lawrence Livermore National Labouratory in the US, remains the world's fastest supercomputer. Frontier at Oak Ridge National Labouratory is in second place and Aurora at Argonne National Labouratory is in third. All three systems are exascale-class machines built and delivered by HPE.

HPE said this marks the third occasion on which systems it built have swept the top three TOP500 positions. A system from the company has now occupied the number one slot for eight consecutive lists.

The results underline a concentration of high-end computing infrastructure around HPE's Cray-derived designs and its liquid-cooled architectures. The company is rolling out a second generation of exascale systems and a refreshed range of hardware and software products.

"HPE continues to advance the next era of AI and high-performance computing through a powerful combination of portfolio innovation and open-source leadership, delivering impact across industries," said Khai Peng Loh, Vice President and Managing Director - Singapore and Southeast Asia, HPE. "Our leadership in building the world's three fastest exascale systems and supporting six of the top ten supercomputers globally highlights the importance of scalable, energy-efficient supercomputing as organizations look to accelerate AI-driven discovery and innovation with tangible societal benefits."

El Capitan delivers 1.809 exaflops of performance and has extended its lead with a 4% gain over previous benchmarks. The machine continues to hold the top position on the High-Performance Conjugate Gradient benchmark. It ranks 23rd on the Green500 index of energy efficiency and runs on AMD Instinct MI300A accelerated processing units.

Frontier, hosted at Oak Ridge, sustains 1.353 exaflops. The system uses AMD Instinct MI250 graphics processors alongside AMD Epyc central processors. It holds third place on the HPL-MxP mixed-precision benchmark for combined high-performance computing and AI workloads with 11.4 exaflops.

Aurora at Argonne National Labouratory delivers 1.021 exaflops in double precision. It ranks second on HPL-MxP with 11.6 exaflops for mixed-precision calculations. HPE built Aurora in collaboration with Intel.

Three further HPE-built systems sit in the TOP500 top ten. HPC6, owned by Italian energy group Eni, is ranked sixth with 477.9 petaflops and is described as the most powerful on-premises industrial supercomputer. Eni uses HPC6 for industrial process optimisation, fluid dynamics analysis for carbon dioxide storage, and plasma simulations for magnetic confinement fusion.

Alps at the Swiss National Supercomputing Centre is ranked eighth with 434.9 petaflops. It serves researchers in Switzerland and internationally in areas such as medicine, quantum chemistry, climate science and machine learning. LUMI, a EuroHPC system based in Finland, is ninth with 379.7 petaflops and supports work in climate modelling, neural network training in European languages and a range of other disciplines.

Energy rankings

On the Green500 list of energy-efficient supercomputers, ten of the top 20 systems were built and delivered by HPE. The machines use fanless direct liquid cooling derived from Cray designs.

Isambard-AI at the University of Bristol is ranked fourth on the Green500. The system forms part of the UK government's AI Research Resource and underpins research projects in robotics, data analysis, climate science and drug discovery.

Samsung Electronics' SSC-24 Energy Module is seventh on the Green500 and 21st on the TOP500. HPE said SSC-24 is the most energy-efficient supercomputer owned by a commercial enterprise. Helios GPU at Academic Computer Centre Cyfronet AGH in Poland is eighth on the Green500 and 96th on the TOP500. It is currently the fastest system in Poland.

Portage, a benchmarking system that HPE operates for its own testing of high-performance and AI workloads, ranks 10th on the Green500 and 70th on the TOP500. Hunter at the High-Performance Computing Centre of the University of Stuttgart is 12th on the Green500 and supports engineering and applied science for industrial and academic users.

Three systems linked to Lawrence Livermore and Oak Ridge appear in the Green500's second ten. RZAdams, a sibling to El Capitan, is 16th on the energy list and 72nd on the TOP500. Frontier TDS, a test and development system for Oak Ridge, ranks 17th on the Green500 and also features in the TOP500 top 100. Tuolumne, an open science companion to El Capitan, ranks 19th on the Green500 and 12th on the TOP500.

Shaheen III - GPU at King Abdullah University of Science and Technology in Saudi Arabia sits in 18th position on the Green500 and 18th on the TOP500. It is currently the most powerful supercomputer in the Middle East.

Sovereign AI systems

HPE has also reported a series of new high-performance installations that focus on national or regional AI infrastructure. These systems have appeared for the first time in recent TOP500 and Green500 rankings.

The TELUS sovereign AI factory in Canada launched in September and delivers 22.74 petaflops. It is now the fastest supercomputer in the country. The system supports Canadian businesses, researchers and public sector bodies that require domestic hosting for AI workloads and data.

Alem.Cloud Sovereign AI in Kazakhstan has entered the TOP500 at number 86 with 20.48 petaflops. The system operates in Astana under the Ministry of Artificial Intelligence of the Republic of Kazakhstan. It is the country's most powerful computer and backs government services, universities, start-ups and local IT firms.

Other recent contract wins for HPE include systems named Discovery and Lux, Mission and Vision, Janus, and a machine focused on quantum error correction for hybrid quantum-classical research.

Research workload

HPE said its supercomputers now underpin a wide range of scientific projects. These span climate modelling, materials science, biology, energy research and engineering.

Researchers using El Capitan have developed high-fidelity simulations for protein folding that incorporate deep learning methods. The system has also generated large libraries of earthquake and tsunami simulations that form a detailed digital model of the seafloor. Scientists are using these data sets in work on tsunami prediction and coastal early-warning approaches.

Frontier and Alps have supported large-scale earthquake simulations for civil engineering. A team at ETH Zurich has used the systems for what it describes as the first full quantum mechanical simulation of a nanoribbon transistor. Researchers linked to the Climate Gordon Bell Prize have run high-resolution global climate models on Alps at 1.25 kilometre grid spacing and have been able to reach 1 kilometre resolution in parts of the climate system.

Aurora is working with the Advanced Photon Source X-ray facility. The project is building data libraries for research into battery materials at atomic scale. The aim is longer battery life and more efficient energy storage.

European researchers using LUMI and other EuroHPC systems have recently achieved multi-decade, fully coupled global climate simulations at about 5 kilometre resolution and have scaled to 1 kilometre resolution across all parts of the simulated climate. The work supports analysis of cloud formation and extreme weather.

Open-source software

HPE's long-standing link with the Chapel programming language also advanced this year. Chapel has joined the High Performance Software Foundation as its latest project. Cray originally created Chapel to simplify programming on parallel systems. HPE has acted as steward for the project in recent years.

Chapel will now sit under community governance as part of the Linux Foundation structure. HPE will continue as a founding member and contributor as the language seeks a broader user and developer base.

The company is continuing to deploy second-generation exascale designs, new liquid-cooled architectures and open-source tools as national laboratories, universities and enterprises plan their next rounds of AI and high-performance computing investment.