NVIDIA’s Role in Advancing Supercomputing: A New Era of Speed and Science

Modern high-performance computing (HPC) is transcending its traditional role of quick calculations to fuel artificial intelligence (AI) systems that are unlocking significant scientific breakthroughs. According to a recent article from NVIDIA, the company is at the forefront of this evolution, powering an impressive 77% of the systems on the latest TOP500 list of the world’s most powerful supercomputers, unveiled at ISC 2025.

NVIDIA Dominates the TOP500 and Green500 Lists

NVIDIA’s influence in the supercomputing arena is underscored by its presence in 381 systems on the TOP500 list, including the Jülich Supercomputing Centre’s JUPITER, which debuted at #4. Notably, 83 of the top 100 systems are now leveraging accelerated computing, while only 17 rely solely on CPUs. Furthermore, NVIDIA’s GH200 Grace Hopper Superchips power the top three systems on the Green500 list, highlighting their energy efficiency.

Tensor Cores: Revolutionizing AI and Scientific Discovery

Central to NVIDIA’s impact are its Tensor Cores, which expedite matrix operations essential for AI and deep learning. These components allow for faster computations, especially as organizations transition to lower precisions like FP8 for model training. NVIDIA is actively developing libraries to enhance high-precision tensor and matrix calculations, focusing on accuracy, performance, and energy efficiency.

A study by Yuki Uchino and Katsuhisa Ozaki demonstrated the potential of using Integer Matrix Multiply Accelerators in Tensor Cores for arbitrary precision, including FP64. Inspired by this, NVIDIA is creating libraries that have already shown significant benefits, such as a 1.8x speed increase in silicon simulations, conserving both time and energy.

AI Supercomputing: Driving Scientific Progress

The capabilities of mixed precision and AI in supercomputing are leading to significant scientific advancements. Noteworthy achievements include the Nobel Prizes awarded to researchers using AI for protein structure prediction, and the Gordon Bell Prize for high-performance computing awarded to David Keyes’s team for their work on the ERA5 climate dataset.

The University of Bristol’s Isambard-AI system, powered by NVIDIA Grace Hopper, exemplifies the use of mixed precision in training models like Nightingale, a multimodal foundation model for healthcare and biomedical research. This model integrates various data types to provide medical insights, demonstrating the potential of AI in advancing scientific workflows.

The Future of HPC

The integration of accelerated computing, tensor technologies, and mixed precision methods is reshaping computational science, paving the way for more AI-driven breakthroughs. With systems like JUPITER and innovations such as the Ozaki emulation method, the future of supercomputing is poised for transformative developments.

As the field evolves, the focus will shift to smart, flexible approaches that prioritize scientific discovery without compromising integrity, meeting the needs of the scientific and HPC communities globally.

For more information, visit the NVIDIA blog.

Image source: Shutterstock