Nine months after the unveiling of Nvidia’s Grace Hopper Superchip, the company announced at the ISC High Performance event in Hamburg, Germany this week that nine new supercomputers around the world are using the chip.
Nvidia also took ISC as an opportunity to discuss its progress in helping individual nations around the world set up their own “sovereign AI” programs. In addition, it shed light on how it is working with quantum computing partners to enable hybrid quantum-classical accelerated supercomputing at three major supercomputer sites.
Among the nine new supercomputers using the Grace Hopper Superchip, Nvidia identified:
-
EXA1-HE, in France, from CEA and Eviden, delivered last month
-
Helios at Academic Computer Centre Cyfronet, in Poland
-
Alps at the Swiss National Supercomputing Centre from Hewlett-Packard Enterprise (HPE)
-
JUPITER at the Jülich Supercomputing Centre in Germany
-
DeltaAI at the National Center for Supercomputing Applications at the University of Illinois Urbana-Champaign
-
Miyabi at Japan’s Joint Center for Advanced High Performance Computing — established between the Center for Computational Sciences at the University of Tsukuba and the Information Technology Center at the University of Tokyo.
Dion Harris, head of data center product marketing at Nvidia, said these supercomputers and others using Nvidia chips are powering AI processing that is being used to accelerate scientific research in areas like climate change, drug discovery, and more, leveraging “this novel sort of architecture of a tightly coupled CPU and GPU in the same architecture to deliver great performance for HPC and AI.”
Harris added that 200 exaflops worth of Grace Hopper-based AI computing power will be coming online just in 2024. The Alps supercomputer in Switzerland alone will feature 20 exaflops of AI, driven by 10,000 Grace Hopper Superchips. It will be the “fastest supercomputer in Europe,” he said, and also will be 10x more energy efficient than Piz Daint, another Swiss supercomputer, as it tackles HPC and AI applications related to research around weather, climate models, and material science.
In addition, Isambard-AI and Isambard 3 from the University of Bristol in the U.K. and systems at the Los Alamos National Laboratory and the Texas Advanced Computing Center in the U.S. join a growing wave of Nvidia Arm-based supercomputers using Grace CPU Superchips and the Grace Hopper platform.
The Isambard-AI machine is emblematic of the “sovereign AI” movement, a recognition by individual nations around the world that AI can be of strategic and cultural importance to them, motivating them to own and host more of their own AI-related data, infrastructure, and workforces to foster innovation. The Gefion supercomputer in Denmark, which has 1,528 Nvidia H100 chips, and the Jean Zay supercomputer in France, which has 1,456 H100s, also fit with the sovereign AI mission.
Isambard-AI phase one consists of a HPE Cray Supercomputing EX2500 with 168 Nvidia GH200 Superchips, making it one of the most efficient supercomputers ever built, Harris said. When the remaining 5,280 NVIDIA Grace Hopper Superchips arrive at the University of Bristol’s National Composites Centre this summer, it will increase performance by about 32x.
“Isambard-AI positions the U.K. as a global leader in AI, and will help foster open science innovation both domestically and internationally,” said Prof. Simon McIntosh-Smith, University of Bristol. “Working with Nvidia we have delivered phase one of the project in record time, and when completed this summer will see a massive jump in performance to advance data analytics, drug discovery, climate research and many more areas.”
Similar to sovereign AI, there are also numerous nation quantum computing initiates around the world–more than 25 by Nvidia’s count–which are feeding the nascent trend toward “quantum-integrated supercomputing.”
The ABCI-Q supercomputer currently being built in Japan for the Advanced Industrial Science and Technology (AIST) with 2,020 H100 chips and the previously mentioned Gefion in Denmark both are leveraging Nvidia’s CUDA-Q programming platform to help drive quantum research being done on those machines. At ISC, Nvidia announced that CUDA-Q also will play a key role in pairing three supercomputers–Japan’s ABCI-Q, Germany’s JUPITER, and the Poznan Supercomputing and Networking Center in Poland–with quantum processing units from quantum computing vendors–QuEra Computing in Japan, IQM Quantum Computers in Germany, and Orca Computing in Poland.
Tim Costa, director of quantum and HPC at Nvidia, said that in each of these cases the GPU-powered supercomputer will handle most computations, but will hand off key tasks to the quantum computer alongside it. Meanwhile, each quantum computer will benefit from co-location with a classical supercomputer that will help its need for calibration, control, and error correction, he said.
“Useful quantum computing will be enabled by the tight integration of quantum with GPU supercomputing,” Costa added. “Nvidia’s quantum computing platform equips pioneers such as AIST, JSCand PSNC to push the boundaries of scientific discovery and advance the state of the art in quantum-integrated supercomputing.”