NVIDIA has announced that a range of the world’s leading computer manufacturers are adopting new NVIDIA Grace superchips to create the next generation of servers that turbocharge AI and HPC workloads for the exascale era. Atos, Dell Technologies, GIGABYTE, HPE, Inspur, Lenovo and Supermicro plan to deploy servers built with NVIDIA Grace CPU Superchip and NVIDIA Grace Hopper Superchip chips.
Grace Superchip CPU
All of these new systems benefit from the just-announced Grace and Grace Hopper designs in the NVIDIA HGX platform, which provide manufacturers with the blueprints needed to build systems that deliver the best performance and twice the memory bandwidth and power. energy efficiency of today’s leading data center processors.
As supercomputing enters the era of exascale AI, NVIDIA is teaming up with its OEM partners to enable researchers to tackle colossal challenges previously unattainable. In the fields of climatology, energy research, space exploration, computational biology, quantum computing and more, the NVIDIA Grace CPU Superchip and Grace Hopper Superchip form the basis of the platform. world’s most advanced form of HPC and AI.
—Ian Buck, VP Hyperscale and HPC at NVIDIA
Major supercomputing centers in the United States and Europe will be among the first to have systems with superchips.
Los Alamos National Laboratory has announced that Venado, its next-generation system, will be the first system in the United States to be powered by NVIDIA Grace CPU technology. Built using the HPE Cray EX supercomputer, Venado is a heterogeneous system that will include a mix of Grace CPU Superchip nodes and Grace Hopper Superchip nodes for a broad and emerging set of applications. Upon completion, the system should exceed 10 exaflops of AI performance.
Alps, the new system of the Swiss National Computing Center, which will also be built by HPE using the HPE Cray EX supercomputer, will use the Grace CPU chip to enable groundbreaking research in a wide range of fields. It will serve as a generalist system open to the scientific community in Switzerland, as well as to the rest of the world.
NVIDIA Grace accelerates compute-intensive workloads. The NVIDIA Grace CPU superchip features two Arm-based processors, cohesively connected via the high-bandwidth, low-latency, and low-power NVIDIA NVLink-C2C interconnect. This design features up to 144 high-performance Arm Neoverse cores with scalable vector extensions and a 1 terabyte per second memory subsystem.
The Superchip Grace CPU interfaces with the latest PCIe Gen5 protocol to enable maximum connectivity with the highest performance GPUs, as well as NVIDIA ConnectX-7 intelligent network interface cards and NVIDIA BlueField-3 DPUs for loads secure HPC and AI workspaces.
The Grace Hopper Superchip combines an NVIDIA Hopper GPU with an NVIDIA Grace processor in an integrated module connected to NVLink-C2C to address large-scale HPC and AI applications.
Systems powered by NVIDIA Grace will run the NVIDIA AI and NVIDIA HPC software portfolio for full-stack embedded computing.