Not even a week after Nvidia announced that it would be providing hardware for the first four of the eight planned EuroHPC systems, HPE and AMD are announcing another major EuroHPC design win. Finnish supercomputing center CSC revealed today that its forthcoming LUMI system – commissioned under the auspices of EuroHPC – will be built by HPE using AMD processor technology. With a target Linpack performance of 375 petaflops, LUMI is the second of three planned “pre-exascale” machines under EuroHPC to have its hardware detailed. The system also breaks new ground in energy efficiency.
LUMI is based on the HPE Cray EX supercomputer architecture, and will harness next-generation AMD Epyc CPUs and AMD Instinct GPUs. Storage will include 7 PB of accelerated flash-based storage (LUMI-F, utilizing a Cray ClusterStor E1000 storage system); an 80 PB Lustre file system (LUMI-P); and 30 PB of encrypted object storage (LUMI-O). LUMI’s primary GPU-driven partition (LUMI-G) will be supplemented by a data analytics partition with 32 TB of memory and additional GPUs (LUMI-D), as well as a CPU partition featuring around 200,000 AMD Epyc CPU cores. LUMI will use HPE Slingshot networking.
LUMI’s main partition is expected to deliver over 550 peak petaflops of computing power, just above (current Top500 leader) Fugaku’s 513.8 peak petaflops. Committed Linpack performance, meanwhile, is 375 petaflops (shy of Fugaku’s 415.5). This makes LUMI the most powerful of the five planned EuroHPC systems that have been detailed so far: Italy’s Leonardo system is expected to deliver over 200 peak petaflops; Luxembourg’s Meluxina, over 18; the Czech Republic’s EURO IT4I, 15.2; and Slovenia’s Vega, 6.8.
Energy efficiency & siting
LUMI will support research from ten countries, comprising a new “LUMI consortium”: Belgium, the Czech Republic, Denmark, Estonia, Finland, Iceland, Norway, Poland, Sweden and Switzerland. The consortium says that LUMI will enable more precise climate models; advanced AI applications in areas like self-driving vehicle development; large-scale social science analytics; personalized medicine research; and much more. Up to 20 percent of LUMI’s capacity will be reserved for European industry and small- and medium-sized enterprises (SMEs).
The system will be installed in a new datacenter at CSC’s campus in Kajaani, Finland, where CSC says the system will take up roughly the size of a tennis court. The datacenter is expected to be ready by Q4 2020; the first phase of LUMI, by Q2 2021; and the second phase of LUMI, by Q4 2021. LUMI will use 100 percent renewable energy (thanks to local hydropower) and its waste heat will be used to supply approximately 20 percent of the yearly district heating needs of its host town, resulting in a stated net negative carbon footprint of 13,500 tons of CO2 equivalent per year.
HPE says that LUMI will use “approximately 8.5 megawatts,” which translates into an extremely efficient 44.1 gigaflops per watt. For context, this is nearly double the efficiency of the current Green500 leader, MN-3, which delivers 21.1 gigaflops per watt. Extrapolating LUMI’s numbers out, an exascale system would require just around 22.7 megawatts — well within the 40-megawatt limit targeted by the planned U.S. exascale systems (a goal that requires a minimum efficiency of 25 gigaflops per watt), and close to the ambitious 20-megawatt exascale goal set by the U.S. Defense Advanced Research Project Agency (DARPA) in the early 2010s. With HPE supplying those three exascale systems as well, this may be the first real preview of those systems’ computational efficiency.
“Once operational in mid-2021, the LUMI supercomputer will be one of the most competitive and green supercomputers in the world,” said Anders Jensen, executive director of EuroHPC. “Such [a] leadership-class system will support European researchers, industry and [the] public sector in better understanding and responding to complex challenges and transforming them into innovation opportunities in sectors like health, weather forecasting or urban and rural planning.”
LUMI’s total cost of ownership through 2026 is expected to be approximately €200 million ($237.1 million), of which €140 million ($165 million) will go toward the supercomputer itself. This cost will be split between the European Commission (€100 million), Finland (€50 million) and the remaining nine countries in the LUMI consortium (€50 million).
The EuroHPC roadmap
LUMI is the second EuroHPC system so far that is being built by HPE (with the other being EURO IT4I). Leonardo, Meluxina and Vega, meanwhile, are all being built by French IT firm Atos. As a U.S. company, HPE appears to have successfully wooed EuroHPC with promises to bring some of its operations across the pond: HPE will be manufacturing liquid-cooled Cray EX supercomputers and Apollo systems in its Kutná Hora factory in the Czech Republic (where EURO IT4I will be based), and the company has also committed to establishing a Center of Excellence in Europe that will provide R&D services and expertise in support of exascale readiness.
Three of the eight planned EuroHPC systems are yet to be detailed: the four-petaflops PetaSC system, hosted by Sofia Tech Park in Bulgaria; the ten-petaflops Deucalion system, hosted by the Minho Advanced Computing Center in Portugal; and the 200-petaflops MareNostrum 5 system, hosted by the Barcelona Supercomputing Center in Spain. By process of elimination, MareNostrum 5 will be the last of the three pre-exascale systems to be fully announced. “For the third machine, the tendering process is in its very final phase now,” Jensen said, “so there will be announcements on that in the coming weeks.”
LUMI is the latest in a series of wins for HPE’s Cray EX systems, which will serve as the basis for the three planned U.S. exascale systems (Aurora, El Capitan and Frontier) and Australia’s most powerful supercomputer (just announced this week). Conspicuously absent, however, are mentions of the European Processor Initiative (EPI) and its planned first-generation “Rhea” CPUs, ostensibly slated for 2021. Jensen maintains that the EPI continues to factor into the future of EuroHPC.
“The ambition is that when we get to the exascale systems – which is what we will be discussing in the coming years – … that EPI will deliver a processor that we can base at least one of the upcoming exascale systems on,” he said.
Header image: concept art of the LUMI supercomputer. Image courtesy of CSC.