Nvidia recently announced its Grace Hopper GH200 Superchip with its own Arm-based CPU and Hopper-based GPU for artificial intelligence (AI) and high-performance computing (HPC). Although the Nvidia H200 is very similar to the previous H100, it makes a big difference with the new HBM3e memories. In addition to this hardware, the company also announced the supercomputer called JUPITER. JUPITER will be very important for Europe.
Europe’s first ExaFLOPS supercomputer
JUPITER, Europe’s first ExaFLOPS supercomputer, hosted at the Forschungszentrum Jülich facility in Germany, is built on Nvidia’s Grace Hopper GH200 Superchips. The machine can be used for both simulations and AI workloads, setting it apart from the vast majority of supercomputers installed today. Additionally, Nvidia said the GH200 powers 40 supercomputers worldwide and their combined performance is around 200 ‘AI’ ExaFLOPS.
The Jupiter supercomputer is powered by approximately 24,000 GH200 chips with a total of 2.3TB of HBM3E memory, connected together using the Quantum-2 InfiniBand network and cooled using Eviden’s BullSequana XH3000 liquid cooling technology. The machine delivers 1 ExaFLOPS FP64 performance for simulations such as climate and weather modelling, materials science, drug discovery, industrial engineering and quantum computing. In addition, the machine can deliver 90 ExaFLOPS AI performance for training large language models and similar workloads. All of this will be supported by Nvidia’s software solutions such as Earth-2, BioNeMo, Clara, cuQuantum, Modulus and Omniverse.Each node on the Jupiter supercomputer contains 288 Arm Neoverse cores and four H200 GPUs for AI and HPC workloads, delivering an incredible AI performance of 16 PetaFLOPS. The GH200 superchips are interconnected using Nvidia’s NVLink interconnect, but Nvidia hasn’t disclosed the total bandwidth of this interconnect.
While Jupiter is a remarkable system – at the end of the day, it’s Europe’s first ExaFLOPS machine – it’s not the only GH200-based supercomputer. Notably, the University of Bristol is building a supercomputer that aims to be the UK’s most powerful, with more than 5,000 GH200s with 141GB HBM3E. Nvidia’s Grace Hopper GH 200 will be used in more than 40 supercomputers built worldwide by companies like Dell, HPE and Lenovo. The combined performance of these systems will be approximately 200 AI ExaFLOPS.
50700
