Nvidia and the German research center Jülich have the key data Jupiter announced: 23,752 GH200s make the supercomputer the fastest in Europe. There is a lot of performance not only for artificial intelligence (AI), the ExaFLOPs brand also applies to HPC applications. And all this with just 18.2 megawatts.
Very fast and efficient
Climbing higher, faster and further without having to look at your electricity bill (and purchase costs) is very easy. But if power consumption doesn’t get completely out of control, this automatically limits your hardware choice.
Germany’s fastest supercomputer, which is also the new number 1 in Europe, can also be sold better politically by focusing on efficiency; after all, (industrial) electricity and its price are a current problem in this country. And since the public sector massively supports the Jupiter project, so it must be: 250 million euros will come from the European supercomputing initiative EuroHPC JU and 250 million euros will come equally from the Federal Ministry of Education and Research (BMBF) and the Ministry of Culture and Science of the Land of North Rhine-Westphalia (MKW NRW).
18.2 megawatts of power consumption at 1.0 ExaFPLOPS HPC performance is in the league where the current #1 Frontier system runs on AMD hardware. According to the May 2023 Top500, Frontier has so far delivered 1.2 ExaFLOPS of performance at just over 20 megawatts. Recently, however, Frontier has been further optimized; for a perfect comparison with Jupiter we still have to wait for new data. Once again Frontier and Jupiter make it clear one way or another: Without a GPU, nothing works in supercomputers.
GH200 at its best
The GH200 aka Grace Hopper Superchip is used by the tens of thousands in the JUPITER supercomputer, the abbreviation stands for “Pioneering joint venture for innovative and transformative exascale research“, installed, more precisely in 23,752 versions. GPUs find their place in the so-called booster module.
For this purpose, Nvidia has built a new quad pack for the more efficient and space-saving use of four GH200s, which Eviden (part of the Atos group) will then install in the Atos BullSequana XH3000. A quad pack marks one node, two nodes are installed in one blade – this has recently been the rule with these solutions. Of course the blade is liquid cooled, which is no longer possible with these packaged products that consume a lot of electricity.
The entire system will occupy the space of approximately 4 tennis courts and will use over 260km of high-performance cables
The cluster module will be equipped with SiPearl’s new Rhea processor designed and manufactured in Europe, a CPU with high memory bandwidth for complex workloads. The cluster and booster modules are dynamically managed as a unified supercomputer using ParTec’s modular ParaStation Module operating system.
Installation from early 2024
Installation of Jupiter will begin in early 2024. With construction, scientific users will have the opportunity to prepare and test the system as part of the early access program. This allows for close collaboration between all parties involved to produce and configure the best possible version of the system for the scientific community, the partners explained in a press release.
Update November 14, 2023 9:52 am
ComputerBase received information about this item from Nvidia under NDA. The only requirement was the earliest possible publication date.