NVIDIA has unveiled a supercomputer based on the NVIDIA Grace CPU Superchip as part of a recent wave of energy-efficient supercomputers using the Arm Neoverse architecture.



The company has collaborated with HPE, British academics and other parties to develop this powerful computing equipment in a bid to compete with AMD and Intel in the semiconductor industry.

The Isambard 3 supercomputer is built with 384 Arm-based NVIDIA Grace CPU Superchips, situated in the Bristol and Bath Science Park in the UK, powering medical and scientific research. The system is expected to deliver six times the performance and energy efficiency of the previous iteration, making it one of Europe's most energy-efficient supercomputers.

The supercomputer is also one of the top three most environmentally friendly non-accelerated supercomputers worldwide, with a peak performance of approximately 2.7 petaflops on the FP64 benchmark and power consumption of under 270 kilowatts. The University of Bristol, in conjunction with the universities of Bath, Cardiff, and Exeter, is leading the project as part of the GW4 Alliance.

Isambard 3's incredible application performance efficiency of up to 6x its predecessor will provide scientists with a revolutionary new supercomputing platform to advance groundbreaking research. The system will help accelerate advancements in AI, life sciences, medicine, astronomy, biotech, and make intricate simulations of renewable energy systems and fusion reactors.

The Arm-based, NVIDIA Grace-powered system will focus on the simulation of molecular-level processes to better understand Parkinson's disease and discover fresh approaches to treating COVID-19 and osteoporosis. The best-in-class cores, memory bandwidth, and memory capacity per core of Grace's architecture make it well equipped for computationally demanding applications.

Simon McIntosh-Smith, HPC Professor at the University of Bristol and lead investigator for the Isambard project, believes that Isambard 3 will provide scientists with a revolutionary new supercomputing platform to advance groundbreaking research. The university expects the current 800 registered users to more than double after the system goes live in spring 2024.

NVIDIA's expertise in high-performance computing, particularly in the fields of graphics processing units (GPUs) and deep learning, meets Arm's energy-efficient architecture to form a powerful combination. This collaboration aims to optimize performance and power efficiency across a wide range of workloads, from cloud services and data centers to edge computing devices.

At the core of this partnership is the goal of delivering exceptional performance while minimizing power consumption. By harnessing the computational power of NVIDIA GPUs and the energy efficiency of Arm-based processors, the Neoverse ecosystem is poised to redefine the possibilities of efficient and sustainable computing.

The implications of this collaboration are far-reaching. The potential for enhanced energy efficiency in AI, machine learning, and other compute-intensive workloads opens new frontiers in various industries, including healthcare, automotive, scientific research, and more. Furthermore, by driving advancements in energy-efficient computing, NVIDIA and Arm are positioning themselves at the forefront of the movement towards sustainable technology.
The NVIDIA and Arm Neoverse ecosystem represents a pioneering force in the realm of energy-efficient computing, promising not only superior performance and reduced power consumption but also a brighter, more sustainable future for technology. This partnership showcases a commitment to innovation that transcends traditional boundaries, making strides towards a greener and more powerful world of computing.