The 2024 edition of the ISC High Performance Computing (HPC) Show took place this week in Hamburg, Germany. Given Nvidia’s role in HPC (HPC), they have been a staple at the event for years. At this year’s event, the company made several significant announcements.

First, looking to support everything from climate and weather to scientific exploration, the company announced it brought the Arm-based Nvidia Grace CPU and Hopper GPU together using its NVLink-C2C interconnect technology. Together, they produce 200 exaflops—the equivalent of 200 quintillion calculations per second. Second, Nvidiasaid that supercomputers in Germany, Japan, and Poland will use its CUDA-Q Platform to accelerate quantum computing efforts.

NVLink was a major advancement in accelerated computing as it enables CPUs and GPUs be to be connected and creates a system that acts as a single chip. This has enabled Nvidiato push the limits of accelerated computing and create use cases, such as the ones highlighted at ISC24.

I had the opportunity to attend a briefing before the announcement with two executives from Nvidia — Dion Harris, director of data center product solutions, and Timothy Costa, director of HPC and quantum computing.

Grace Hopper: powering supercomputers

Countries around the globe are looking to create “sovereign AI”—hosted and owned systems, with infrastructure and data all in-country. Grace Hopper can serve as the scientific supercomputing engine for those countries.

“It leverages the novel architecture of a tightly coupled CPU and GPU to deliver great performance for HPC and AI,” Harris said. “One of the other key drivers for adopting Grace Hopper is energy efficiency. It can be about 2x more energy efficient on average compared to an x86 plus a GPU config.”

Harris said Alps, the first European Grace Hopper supercomputer, is now online. He added that it’s the fastest AI supercomputer in Europe, with 20 Exaflops of AI. Also, it’s ten times more energy efficient than Piz Daint, its predecessor supercomputer. Alps is powered by 10,000 Grace Hopper Superchips and can support HPC and AI with a focus on weather, climate, and material science.

“The key use of the system will be to drive innovation around weather and climate in particular, trying to generate high-quality, high-fidelity one-kilometer scale global climate simulation models,” Harris said. “And then they can, in turn, take those simulation models and use them to train and build advanced AI surrogate models. So you have this feedback loop that will also accelerate science for climate technology.”

CUDA-Q platform provides national supercomputing centers with open-source access

Nvidia also announced that it’s working with national supercomputing centers worldwide to utilize the open-source Nvidia CUDA-Q platform. Supercomputing sites in Germany, Japan, and Poland will use the platform to power the quantum processing units (QPUs) inside their Nvidia-accelerated HPC systems.

“Quantum accelerated supercomputing, in which quantum processors are integrated into accelerated supercomputers, represents a tremendous opportunity to solve scientific challenges that may otherwise be out of reach,” Costa said. “But there are several challenges between us today and useful quantum accelerated supercomputing.”

He noted that qubits are noisy and error-prone, and integration with HPC systems, error correction, infrastructure, and exponential-speed algorithms are still on the to-do list. NVIDIA says that supercomputers in Germany, Japan, and Poland will use the company’s CUDA-Q platform, which should remove those barriers that Costa mentioned.

“At AIST in Japan, a QuEra Neutral-Atom quantum processor will be coupled with the ABCI-Q supercomputer,” Costa said. “At Jülich in Germany, an IQM superconducting quantum processor will be coupled to the JUPITER Grace Hopper supercomputer. And at PSNC in Poland, two ORCA photonic quantum processors have arrived and will be coupled to their H100 supercomputer.”

Getting real about AI, HSC and supercomputing

Jensen Huang, CEO of Nvidia, has been an AI visionary and has steadily brought to market the tech that makes it possible. He talked about AI and building the tools long before it became fodder for the media. As evidenced by announcements like these, the company is making pipedreams possible.

Grace Hopper is a great leap forward for supercomputing centers worldwide and shows that the company needs to think more unconventionally. The CUDA-Q platform will help supercomputers around the globe solve previously thorny challenges.

I recall that early in the accelerated computing growth curve, Huang would often talk about the future in a way that seemed far-fetched and was the stuff of science fiction. The company has used the impossible as a lighthouse, continuing to reimagine computing and making the right acquisitions and innovations to move ahead in leaps instead of steps. I’ve often been asked by industry watchers and investors whether rivals like Intel can make a dent and take the shine off the darling of the AI industry, and the answer is no. Nvidia understands that delivering on AI, supercomputing, quantum, and other forms of accelerated computing requires a systems approach. Everything from the silicon to application software must work together, and that’s the difference. Most think of NVIDIA as a GPU company but they are, in fact, a systems company that’s enabled the world of science fiction to become a reality.