NVIDIA : l’IA accélère la science dans plus de 80 nouveaux systèmes à travers le monde

Across the globe, a revolution is underway in scientific research, fueled by a powerful new instrument: accelerated computing. NVIDIA, a leading force in this transformation, has unveiled at the SC25 conference in St. Louis, Missouri, that over 80 new scientific systems leveraging its accelerated computing platform have been deployed worldwide in the last year. These systems collectively contribute an impressive 4,500 exaflops of AI performance, opening up unprecedented possibilities for discovery in fields ranging from quantum physics to climate research. This represents a significant leap forward, empowering researchers to tackle complex challenges with unprecedented speed and efficiency. This acceleration is driven by NVIDIA’s full-stack accelerated computing platform, which encompasses GPUs, CPUs, DPUs, NICs, and software, including CUDA-X libraries and NVIDIA AI Enterprise. This unified architecture provides the necessary scale and efficiency to advance science sustainably and at an unprecedented pace. From Europe to Asia and North America, the race to build more powerful supercomputers is on, with NVIDIA at the forefront.

The Rise of AI Supercomputers: A Global Perspective

The deployment of these new systems is a testament to the growing importance of AI in scientific research. The Horizon supercomputer at the Texas Advanced Computing Center (TACC), America’s largest academic supercomputer, is a prime example. Slated to come online in 2026, it will be powered by NVIDIA GB200 NVL4 and NVIDIA Vera CPU servers, interconnected with NVIDIA Quantum-X800 InfiniBand networking. This system, boasting 300 petaflops of performance, will be pivotal in accelerating breakthroughs in science and engineering. Across the Atlantic, Europe is also making strides. The Jülich Supercomputing Centre’s JUPITER system, inaugurated in September, has achieved exaflop performance on the HPL benchmark, marking it as Europe’s first exascale computer. Built with 24,000 NVIDIA GH200 Grace Hopper Superchips, JUPITER is already being used for high-resolution global climate simulation, enabling researchers to run global simulations at kilometer-scale resolution. You can learn more about JUPITER here: JUPITER, inaugurated in September. The United States Department of Energy (DOE) is also investing heavily, partnering with NVIDIA to build seven new AI supercomputers at Argonne National Laboratory (ANL) and Los Alamos National Laboratory (LANL). These systems will leverage NVIDIA Blackwell GPUs and networking, enabling advanced AI modeling for science and energy applications. The largest system at ANL, Solstice, will feature 100,000 NVIDIA Blackwell GPUs, reaching a staggering 1,000 exaflops of AI training compute. The DOE also announced a partnership with Lawrence Berkeley National Laboratory for Doudna, a supercomputer launching in 2026 to support over 11,000 researchers.

Key Applications and Scientific Breakthroughs

These supercomputers are not just about raw computing power; they are designed to tackle specific scientific challenges. The Horizon supercomputer at TACC, for instance, will be instrumental in:

• Simulating the mechanics of disease, using AI-enhanced simulations to study viruses.
• Modeling stars and galaxies across the universe, simulating distant galaxies.
• Investigating novel materials at the atomic scale, studying turbulence, solids, and quantum materials.
• Mapping seismic waves to prepare for earthquakes, improving seismic hazard maps.

The application of these technologies extends beyond these examples, with researchers using these systems for drug discovery, climate modeling, and many other fields. The ability to simulate complex systems at unprecedented scales is revolutionizing the scientific process, allowing researchers to test hypotheses, analyze vast datasets, and make discoveries that were previously impossible.

Technical Underpinnings and Future Prospects

The performance gains are driven by NVIDIA’s latest hardware, including the NVIDIA Blackwell GPUs. These cutting-edge GPUs are designed to deliver exceptional AI performance, enabling researchers to train complex models and run simulations faster than ever before. For instance, the Horizon supercomputer will be able to deliver up to 80 exaflops of AI compute at FP4 precision with its 4,000 NVIDIA Blackwell GPUs. The future looks bright for accelerated computing. With the continued advancements in GPU technology and the increasing availability of powerful AI supercomputers, we can expect to see even more groundbreaking discoveries in the years to come. The convergence of AI, high-performance computing, and scientific research is creating a powerful engine for innovation, driving progress across a wide range of disciplines.

Conclusion

NVIDIA’s role in accelerating scientific discovery is undeniable. The global deployment of over 80 new scientific systems, powered by its technology, signifies a pivotal moment in the history of computing and scientific research. As these powerful tools become more accessible, we can anticipate a surge in breakthroughs, transforming our understanding of the world and paving the way for a more innovative future. The relentless pursuit of exascale computing and the integration of AI into scientific workflows are not just trends; they are the new normal.