New Nvidia Supercomputing chips provide ‘speed boost’ for AI

It's been a new break for Nvidia, and it's not because of gaming or bitcoin mining—2023 saw the chipmaker not only participate in the AI ​​revolution, but push it forward, continuing to set new benchmarks with a laundry list of improvements. For computer power.

Ian Buck, NVIDIA's vice president of high-performance computing and hyperscale data center business, made this point today at the company's Supercomputing 2023 event, saying, “Computers are at the foundation of modern civilization and artificial intelligence.”

He notes that CPU performance has grown 10,000 times over the past 20 years, but that “that trend has stalled” as “more overhead… brings disproportionate cost and power gains.”

Buck framed the day's announcements as part of the “New Computer Act.”

Advances in supercomputing chips

Nvidia's upgrade to Hopper's GPU chips, the NVIDIA H200, offers improved performance but also enables cost and energy savings, Buck said, especially in the field of data centers that accelerate industrial design and simulation.

“Nvidi H200 is the first GPU to offer HBM3e – fast and large memory for accelerating generative AI and large language models, and advancing scientific computing for HPC workloads,” the company officially announced the 14GB memory processor. 4.8 terabytes per second, “nearly double the capacity and 2.4x more bandwidth compared to its predecessor.”

The Grace Hopper superchips, which combine Nvidia's Grace and Hopper technology, power supercomputers built by major manufacturers such as Dell and HPI. The GH200 is used in research and heavy computing tasks, delivering a combined 200 exaflops – one quintillion floating point operations per second – of AI performance.

Top computing centers such as Germany's Jülich Supercomputing Center and the University of Bristol are integrating the GH200 into their systems, Nvidia said. Jupiter faces urgent scientific challenges as Europe's first demonstrable supercomputer.

“Jupiter will be the world's most powerful AI supercomputer, powered by 24,000 GH200 superchips, all connected via Nvidia InfiniBand,” Buck said.

Goodbye CPUs, welcome QPUs

Nvidia talked about the development of quantum processing units (QPUs).

“As future supercomputers will include quantum processors, NVIDIA is leading the way in the era of quantum-accelerated supercomputing with an open programming model called CUDA Quantum to facilitate this integration,” Buck said.

Integrating CUDA quantum into its computing platform has allowed researchers to simulate quantum processors, but NVIDIA's contribution to quantum computing is not just theoretical. For example, BASF researchers used CUDA Quantum to pioneer a hybrid quantum-classical method for simulating chemical reactions. This can, for example, protect people from harmful metals. Similarly, researchers at Stony Brook and Brookhaven National Laboratory are using NVIDIA technologies to simulate particle physics that could replace expensive and time-consuming particle accelerator experiments.

Beyond computing hardware, the company also supports AI-driven businesses, providing startups and enterprises with hardware to train their models, he said. Giant companies like OpenAI and Stability AI rely on Nvidia technology. This involvement is reflected in the company's impressive stock growth, which has seen growth of nearly 200% by 2023, confirming its position as one of the leading businesses dominating the SP500.

The company's role as the engine behind many AI businesses, combined with technological advances such as Hopper chips and the NVIDIA H200, puts it at the forefront of the AI ​​revolution. With the stock market reflecting this growth and AI in everyone's crosshairs, NVIDIA is having a big time.

The future seems to be written not only in codes, but also in the silicon of Nvidia's groundbreaking chips.