Europe’s push in the direction of HPC (Excessive-Efficiency Computing) relevancy is an ongoing effort specified by the European Union’s (EU) EuroHPC Joint Unit initiative (opens in new tab). As a part of this program, the previous continent has already deployed its first pre-exascale system, LUMI, which integrates the most recent know-how from AMD in a quantum-ready system that additionally boasts an awe-inspiring carbon-negative design. However LUMI is a stepping stone in the direction of the actual purpose: post-exascale computing. As lined by Computerbase (opens in new tab), that honor is for the stratospheric-defiant JUPITER (Joint Endeavor Pioneer for Progressive and Transformative Exascale Analysis) supercomputer.
JUPITER shall be put in in Jülich’s Supercomputing Centre in Germany, with the EU setting apart a whopping €500 million (~$522 million) for infrastructure, {hardware}, and set up prices alone. The system, anticipated to be operational someday past the 2024 timeframe, would be the continent’s first to surpass the trillion operations per second threshold.
In contrast to LUMI, JUPITER shall be leveraged in the direction of the fields of local weather modeling, supplies engineering, organic simulations, and sustainable power manufacturing analysis whereas leveraging the most recent AI acceleration. Sadly, these are all computed and memory-demanding workloads, which justify the set up’s excessive worth.
There hasn’t been any official affirmation concerning what {hardware} will energy JUPITER. Nonetheless, the press launch – and the present state of HPC environments – place important accountability on the toes of GPU-based accelerators. JUPITER will seemingly take classes from LUMI’s star-based structure, with completely different supercomputing modules being delivered to bear in GPU visualization, alongside a common, CPU-based accelerator cluster, a high-performance GPU cluster; a quantum computing node in addition to cold and warm storage clusters. As well as, JUPITER will dive one step additional towards extra unique computational fashions than LUMI in that it’s going to embrace a node totally devoted to neuromorphic computing.
The similarities to LUMI would possibly indicate that AMD will once more profit from this huge contract – and we would say there are good causes for that, notably concerning energy effectivity. However Intel is not pleased with being left within the rearview mirror concerning supercomputing installations, which now counts an impressively cut up of 5 out of the world’s ten quickest supercomputers and ten out of twenty make the most of AMD’s {hardware}.
One other chip in Intel’s hat of tips may very well be any leverage gained through its announcement of its Silicon Junction initiative. The blue big is pouring an estimated $80 billion into semiconductor manufacturing and R&D capabilities throughout the EU and serves as a concession to the EU’s tried plan at semiconductor independence. With out official affirmation, this stays hypothesis, so take this with an sufficient quantity of NaCl.
One of the spectacular components concerning JUPITER’s announcement is its energy consumption. Whereas the world’s prime supercomputer, Frontier, reaches a median of 19 MW in energy consumption, JUPITER is claimed to common out at simply 15 MW – reducing 22% in energy necessities inside a few years of {hardware} improvement. And it is an virtually 50% energy consumption discount in comparison with the previous world champion within the supercomputing subject, Japan’s Arm-based Fugaku. Put in in 2020, its common energy consumption is round 29 MW whereas providing “solely” 537.21 PFlop/s in peak efficiency – half that of JUPITER. That is equal to a doubling in energy effectivity in half a decade – a necessary metric concerning environmental sustainability if we have ever seen one.
