During ISC 2025, NVIDIA once again solidified its dominance in the supercomputing landscape. According to the company, 77% of the systems on the latest TOP500 list are powered by NVIDIA technology, including the debut of Germany’s JUPITER system at fourth place.
🔝 TOP500 and Green500 Leadership
Among the TOP100 systems, 83 are now GPU-accelerated, while only 17 rely solely on CPUs, signaling a major architectural shift. Furthermore, the top three systems on the Green500 list—ranking the most energy-efficient supercomputers—are all powered by NVIDIA’s GH200 Grace Hopper Superchips.
💡 Tensor Cores Powering AI and Science
At the core of NVIDIA’s dominance is its Tensor Core architecture, optimized for deep learning and matrix-intensive computations. With growing adoption of lower-precision training (like FP8), NVIDIA is also enhancing libraries for high-precision tasks, balancing accuracy and performance.
A recent study by Yuki Uchino and Katsuhisa Ozaki demonstrated that Tensor Cores can even support FP64 using integer-based matrix multipliers. Inspired by this, NVIDIA developed libraries that showed 1.8x speed gains in silicon simulations—saving time and energy.
🧬 AI and Science Converge
AI-powered supercomputing is rapidly advancing scientific discovery. From protein structure prediction that won a Nobel Prize to climate modeling awarded the Gordon Bell Prize, AI is transforming workflows. Bristol’s Isambard-AI system, powered by Grace Hopper, exemplifies how multimodal models like Nightingale are shaping the future of biomedicine.
