NVIDIA data center chip announces milestones for HPC, AI and quantum computing

In addition to GPUs, NVIDIA is trying to expand into other enterprise-level chip markets. Data center server CPUs have always been Intel's domain and are one of the largest chip markets in the world. NVIDIA is also targeting the high-end market for server CPUs. That is HPC, AI and quantum computing.

As early as 2021, NVIDIA announced that it would launch the server processor Grace in early 2023. Grace, based on the Arm architecture, is mainly targeted at high-end computing tasks such as HPC and large-parameter AI applications. Especially in 2021, when AI models have not yet reached tens of billions, hundreds of billions or even trillions of parameters, NVIDIA still announced the Grace and Grace Hooper chips for the arrival of ultra-large-scale parameter AI models.

The time has come to March 2023. NVIDIA CEO Huang Jenxun said at the GTC conference that Grace’s shipment will be postponed to the second half of 2023. However, during the International Supercomputing Conference ISC 2023 in May, NVIDIA announced a Grace-based supercomputer, which is the British university supercomputing system Isambard 3 in cooperation with HPE.

As part of the UK’s GW4 research consortium, the Isambard 3 project is led by the Universities of Bristol, Bath, Cardiff and Exeter. Isambard 3 will be equipped with 384 Grace CPUs, and its performance and energy efficiency are expected to be 6 times that of Isambard 2, making it one of the most energy-efficient systems in Europe. Isambard 3 is located at the Bristol and Bath Science Park in the UK and is used to advance medicine and scientific research.

Simon McIntosh-Smith, chief researcher of the Isambard project and professor of high-performance computing at the University of Bristol, said: Isambard 3’s application performance efficiency is up to 6 times that of its predecessor, comparable to the top 50 supercomputers in the world’s top 500 supercomputers . Isambard 3 is expected to enter service in spring 2024.

NVIDIA also announced that more and more supercomputers around the world are using NVIDIA chips based on Arm architecture. In addition to Isambard 3, the Swiss National Supercomputing Center and the Los Alamos National Laboratory in the United States are also building supercomputers equipped with GPUs.

At the ISC 2023 conference, NVIDIA also announced plans to jointly establish a new laboratory with the Jülich Supercomputing Center (JSC) of the Jülich Research Center (FZJ) in Germany, which will cooperate with ParTec AG in Munich. Together we are developing a classical-quantum supercomputer based on the NVIDIA quantum computing platform.

Currently, JSC operates JUWELS, one of Europe’s most powerful supercomputers, and operates Europe’s first quantum computing infrastructure through JUNIQ. The new laboratory will be operated by FZJ, one of the largest interdisciplinary research centers in Europe, and is part of JUNIQ, which is currently using a JUWELS acceleration system equipped with 3,744 NVIDIA A100 Tensor Core GPUs for quantum computing simulations.

Timothy Costa, director of high-performance computing and quantum computing at NVIDIA, said that the unification of quantum computing and GPU supercomputing is one of the keys to achieving breakthroughs in the scientific field in the future.

In terms of quantum computing, during the ISC 2023 conference, NVIDIA, Rolls-Royce and quantum software company Classiq announced a breakthrough. Using NVIDIA's quantum computing platform, the two companies designed and simulated the world's largest computational fluid dynamics (CFD) quantum computing circuit. Rolls-Royce plans to use new circuits to take advantage of quantum in CFD, using both classical and quantum computing methods to simulate the performance of jet engine designs.

It is worth mentioning that Grace Hopper drives the world's first GPU-accelerated quantum computing system that combines quantum computing with the most advanced classical computing - DGX Quantum. NVIDIA also provides developers with a link between GPU and QPU Open source programming model - NVIDIA CUDA Quantum.

The above is the detailed content of NVIDIA data center chip announces milestones for HPC, AI and quantum computing. For more information, please follow other related articles on the PHP Chinese website!