AMD Strengthens Its Presence in Supercomputing: 4 of the Top 10 Systems Use Its Technology

Supercomputing is once again at the forefront of the race for artificial intelligence. For years, AI discussions have nearly always been centered around models and commercial data centers. But the infrastructure that enables climate simulation, material research, reactor design, scientific model training, or accelerating medical research still has a standout name: HPC, high-performance computing.

AMD reaches the new TOP500 list of June 2026 with a strengthened position. The company states that its technology is present in 191 systems on the TOP500 and Green500 lists, an 11% increase from the previous year, and it powers 4 of the world’s 10 most powerful supercomputers and 4 of the 10 most energy-efficient. Additionally, AMD claims that its EPYC processors and Instinct accelerators are included in 41% of the new entries in this edition.

The official ranking confirms AMD’s influence at the top of the list. The new number one is LineShine, installed at the National Supercomputing Centre in Shenzhen, China. Behind it are two major U.S.-based AMD systems: El Capitan at Lawrence Livermore National Laboratory and Frontier at Oak Ridge. Also featured are HPC7 and HPC6, both of which are Eni systems in Italy, within the top 10.

AMD consolidates its presence in the TOP500 elite

The top 10 offers an interesting market insight. Nvidia continues to have a huge presence in AI accelerators, and several high-performance systems use their GPUs. But AMD has managed to place its architecture in key machines for scientific and industrial supercomputing. In the case of El Capitan, Frontier, HPC7, and HPC6, the combination of EPYC CPUs and Instinct GPUs allows for workloads in simulation, modeling, data analysis, and AI.

El Capitan ranks second with 1,809 PFlop/s of measured Linpack performance and 11.34 million cores. Frontier, which was the symbol of the exascale era for years, is third with 1,353 PFlop/s. HPC7, Eni’s new system, appears in sixth position with 571.5 PFlop/s, while HPC6 sits eighth with 477.9 PFlop/s.

Eni’s case deserves attention because it demonstrates that supercomputing is no longer solely a matter for national laboratories. HPC7 and HPC6 are industrial systems oriented toward energy workloads, advanced simulation, modeling, and analysis. AMD presents them as part of Europe’s ambition to strengthen sovereign AI and HPC capabilities—a vision increasingly echoed in EU technological discourse.

LUMI, operated by EuroHPC and hosted by CSC in Finland, ranks 11th, just outside the top 10, but remains one of Europe’s most significant systems. Its role extends beyond raw power: it is a shared infrastructure for research, AI, scientific simulation, and European industrial projects.

Green500: efficiency as the new power metric

The Green500 list is gaining importance because the bottleneck for AI and HPC isn’t just the chip, but the energy it consumes. Training models, running simulations, and operating exascale systems require enormous amounts of electricity and cooling. In this context, watts per efficiency cease to be just a technical detail and become a strategic criterion.

AMD claims that four systems featuring its technology are among the top 10 on the Green500: Otus, Capella, AMD Ouranos, and Portage. It also states that it powers 56% of the 50 most efficient systems on the list.

There’s an important nuance: “with AMD technology” doesn’t always mean a configuration solely composed of AMD CPUs and GPUs. In some systems, AMD components may be limited to CPUs, with accelerators from other vendors. Modern supercomputing is increasingly heterogeneous, combining different architectures with varying processors, GPUs, interconnects, memory, and cooling methods.

Still, the core message is clear. The next phase of scientific AI cannot rely solely on adding accelerators. Centers must optimize performance per megawatt, reduce operational costs, and justify capacity jumps in increasingly energy, sustainability, and network availability-sensitive public or industrial budgets.

Europe seeks sovereignty through supercomputing and scientific AI

This news also has a European dimension. AMD is involved in several projects presented as infrastructure for technological sovereignty, from LUMI to HPC7 and the upcoming Alice Recoque, France’s future exascale supercomputer.

According to AMD and Eviden, Alice Recoque will be France’s first exascale supercomputer and the second in Europe. It will be led by GENCI, operated by CEA, and financed by EuroHPC JU, the Digital Europe program, and the Jules Verne consortium, with participation from France, the Netherlands, and Greece. The project has a total cost of €554 million and will be based on next-generation AMD EPYC CPUs and AMD Instinct MI430X GPUs.

The system is described as an “AI Factory” blending simulation, data analysis, and artificial intelligence. Its applications will include climate and material modeling, energy, digital twins for personalized medicine, and development of European AI models. AMD and Eviden state that the system will use 94 racks with liquid cooling technology and warm water for improved energy efficiency.

Such projects mark the difference between merely using AI and possessing the capacity to research, train, simulate, and validate independently. Europe has entered late into many commercial aspects of generative AI, but supercomputing remains a pathway to sustain science, industry, and specialized models with greater control over data, infrastructure, and political priorities.

FP64 returns to the center of debate

One of AMD’s most interesting points is its emphasis on double-precision FP64 accuracy. In commercial AI, there’s much talk about FP8, FP4, and low-precision formats because they allow training and inference with higher efficiency. But many scientific workloads cannot afford accumulated numerical errors. Climate, materials, fluids, nuclear fusion, and complex physical systems require high precision.

AMD introduced the Instinct MI430X as a GPU aimed at bridging HPC and AI. The company projects over 200 TFLOPs of native FP64 performance—a figure that, if verified in actual products and systems, would make it highly relevant for simulation, modeling, and science driven by AI.

AMD’s technical argument is that the next generation of scientific AI will rely not just on web or text data but on high-fidelity physical simulations. If models are trained on poor-quality or low-precision scientific data, they will inherit those limitations. Training on more accurate simulations allows better capture of complex phenomena.

Therefore, FP64 is not a relic of the past but a future-oriented component for specific applications. AI that aids in discovering materials, optimizing reactors, accelerating drug development, or modeling climate systems needs reliable data. Often, such data originates in supercomputers designed for precision before text generation.

A race that no longer separates HPC and AI

For years, HPC and AI were viewed as adjacent but distinct markets. The former associated with science, simulation, universities, national labs, and heavy industry; the latter with large models, cloud data centers, and accelerators for training and inference.

That boundary is fading. Modern supercomputers run AI workloads. Scientific models require simulation. Industrial companies use machine learning on HPC-generated data. Labs seek infrastructure capable of training, simulating, visualizing, and analyzing within a single platform.

AMD positions itself right at this convergence. Its EPYC and Instinct products compete in both HPC and AI performance, while projects like El Capitan, Frontier, LUMI, HPC7, and Alice Recoque show that the race is not limited to hyperscalers and generic models.

The June 2026 TOP500 list signals another trend: China reclaims the top spot with LineShine, while the U.S. maintains reference exascale systems, and Europe aims to bolster its capabilities with EuroHPC, Eni, GENCI, and national centers. Supercomputing is once again a geopolitical, scientific, and industrial race.

AMD comes out strengthened from this edition, but competition remains fierce. Nvidia dominates much of the AI narrative, Intel retains presence in key systems, and China’s sovereign projects advance with their own architectures. The key point is that AI infrastructure is no longer just about how many GPUs a company can buy but about which architecture supports science, efficiency, and sovereignty over the long term.

Frequently Asked Questions

How many TOP500 systems use AMD technology?
AMD states that its technology is present in 191 systems on the TOP500 and Green500 lists, an 11% increase over the previous year.

Which top 10 supercomputers use AMD?
In the June 2026 TOP500 list, El Capitan, Frontier, HPC7, and HPC6 appear in the top 10 with AMD technology.

Why is the Green500 list important?
Because it measures energy efficiency. In HPC and AI, performance per watt is increasingly critical due to electricity costs, cooling, and sustainability considerations.

What is AMD Instinct MI430X?
It’s a GPU designed to bridge HPC and AI, projecting over 200 TFLOPs of native FP64 performance.

What role does Europe play in this race?
Europe aims to strengthen its sovereignty in AI and supercomputing through systems like LUMI, HPC7, and the upcoming Alice Recoque, financed through EuroHPC initiatives.

via: amd