[ad_1]
AI processing can take an enormous quantity of computing energy, however by the seems of this newest joint challenge from the Jülich Supercomputing Heart and French computing supplier Eviden, energy is not going to be in brief provide. The 2 firms have signed a deal to construct a brand new information centre to deal with an exascale-class supercomputer that they are saying will likely be able to performing one quintillion floating-point operations per second, or one exaFLOPS in HPC (excessive efficiency computing) output.
The JUPITER supercomputer (in the event you’re questioning, the Joint Enterprise Pioneer for Modern and Transformative Exascale Analysis, which as backronyms go is spectacular) was commissioned by Europe’s supercomputer consortium, EuroHPC JU, (by way of Techspot) in October 2023, and the primary of three deliberate to be constructed will likely be contained within the new facility in Germany, with plans to be operational inside a yr.
JUPITER will likely be powered by a number of Nvidia GH200 Grace Hopper Superchips, will likely be put to work totally on AI coaching and is ready turn out to be the world’s strongest AI system, with a staggering 90 exaFLOPs of efficiency anticipated when set to work on coaching AI fashions.
Constructed round 50 container modules put in over 2,300 sq. meters, the modular method taken to deal with this monster machine is hoped to chop the supply time of the challenge by 50 %, and will imply that future upgrades will likely be simpler to implement over time.
The containers will embrace 20 IT items, 15 energy and 10 logistics modules. The hope is that the JUPITER supercomputer contained inside will likely be operational in beneath a yr and delivered for a a lot decrease value, due to the modular nature of the positioning.
If development goes forward as deliberate, this would be the first exascale-class supercomputer in Europe, nonetheless the title of world’s quickest supercomputer when it comes to uncooked HPC velocity is hotly contested.
The present holder of the title is the Hewlett-Packard Enterprise Frontier, a machine able to 1.102 exaFLOPS, although it could not keep on the prime spot for for much longer. El Capitan is at present beneath development at Lawrence Livermore Nationwide Laboratory in California. It is designed to able to 2 exaFLOPS when accomplished, which is at present estimated to be someday in mid-2024.
Nonetheless, the figures for this new European supercomputer look spectacular, and if this new modular development method is profitable it could function a mannequin for rushing up the implementation of comparable tasks but to start development, just like the proposed exascale system at present deliberate in Edinburgh.
As we transfer right into a world the place big quantities of processing energy turn out to be not only a nicety, however a necessity for AI growth, quantum computing and extra, it looks like development strategies are starting to advance in tandem with the ability of the computer systems themselves.
Now, who’s going to allow us to benchmark one? I have already got some concepts…
[ad_2]
Source link
