My blog
Editor's Picks

HPE, AMD and EuroHPC Partner for Pre-Exascale LUMI Supercomputer

Not even per week after Nvidia announced that it will be offering {hardware} for the primary 4 of the eight deliberate EuroHPC methods, HPE and AMD are asserting one other main EuroHPC design win. Finnish supercomputing middle CSC revealed at this time that its forthcoming LUMI system – commissioned underneath the auspices of EuroHPC – shall be constructed by HPE utilizing AMD processor know-how. With a goal Linpack efficiency of 375 petaflops, LUMI is the second of three deliberate “pre-exascale” machines underneath EuroHPC to have its {hardware} detailed. The system additionally breaks new floor in power effectivity.

The system

LUMI relies on the HPE Cray EX supercomputer structure, and can harness next-generation AMD Epyc CPUs and AMD Intuition GPUs. Storage will embody 7 PB of accelerated flash-based storage (LUMI-F, using 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 major GPU-driven partition (LUMI-G) shall be supplemented by a knowledge analytics partition with 32 TB of reminiscence and extra GPUs (LUMI-D), in addition to a CPU partition that includes round 200,000 AMD Epyc CPU cores. LUMI will use HPE Slingshot networking.

LUMI’s fundamental partition is predicted to ship over 550 peak petaflops of computing energy, simply above (present Top500 chief) Fugaku’s 513.8 peak petaflops. Dedicated Linpack efficiency, in the meantime, is 375 petaflops (shy of Fugaku’s 415.5). This makes LUMI essentially the most highly effective of the 5 deliberate EuroHPC methods which have been detailed up to now: Italy’s Leonardo system is predicted to ship over 200 peak petaflops; Luxembourg’s Meluxina, over 18; the Czech Republic’s EURO IT4I, 15.2; and Slovenia’s Vega, 6.8.

Vitality effectivity & siting

LUMI will assist analysis from ten international locations, comprising a brand new “LUMI consortium”: Belgium, the Czech Republic, Denmark, Estonia, Finland, Iceland, Norway, Poland, Sweden and Switzerland. The consortium says that LUMI will allow extra exact local weather fashions; superior AI purposes in areas like self-driving car improvement; large-scale social science analytics; customized drugs analysis; and way more. As much as 20 % of LUMI’s capability shall be reserved for European trade and small- and medium-sized enterprises (SMEs).

The deliberate design for the LUMI datacenter. Picture courtesy of Synposis Architects Ltd. and Geometria Structure Ltd.

The system shall be put in in a brand new datacenter at CSC’s campus in Kajaani, Finland, the place CSC says the system will take up roughly the scale of a tennis court docket. The datacenter is predicted to be prepared by This fall 2020; the primary section of LUMI, by Q2 2021; and the second section of LUMI, by This fall 2021. LUMI will use one hundred pc renewable power (due to native hydropower) and its waste warmth shall be used to provide roughly 20 % of the yearly district heating wants of its host city, leading to a acknowledged internet destructive carbon footprint of 13,500 tons of CO2 equal per yr.

HPE says that LUMI will use “roughly 8.5 megawatts,” which interprets into a particularly environment friendly 44.1 gigaflops per watt. For context, that is almost double the effectivity of the current Green500 leader, MN-3, which delivers 21.1 gigaflops per watt. Extrapolating LUMI’s numbers out, an exascale system would require simply round 22.7 megawatts — nicely throughout the 40-megawatt restrict focused by the deliberate U.S. exascale methods (a aim that requires a minimal effectivity of 25 gigaflops per watt), and near the formidable 20-megawatt exascale aim set by the U.S. Protection Superior Analysis Challenge Company (DARPA) within the early 2010s. With HPE supplying these three exascale methods as nicely, this can be the primary actual preview of these methods’ computational effectivity.

“As soon as operational in mid-2021, the LUMI supercomputer shall be one of the vital aggressive and inexperienced supercomputers on this planet,” mentioned Anders Jensen, government director of EuroHPC. “Such [a] leadership-class system will assist European researchers, trade and [the] public sector in higher understanding and responding to advanced challenges and reworking them into innovation alternatives in sectors like well being, climate forecasting or city and rural planning.”

LUMI’s whole value of possession by 2026 is predicted to be roughly €200 million ($237.1 million), of which €140 million ($165 million) will go towards the supercomputer itself. This value shall be cut up between the European Fee (€100 million), Finland (€50 million) and the remaining 9 international locations within the LUMI consortium (€50 million).  

The EuroHPC roadmap

LUMI is the second EuroHPC system up to now that’s being constructed by HPE (with the opposite being EURO IT4I). Leonardo, Meluxina and Vega, in the meantime, are all being constructed by French IT agency Atos. As a U.S. firm, HPE seems to have efficiently wooed EuroHPC with guarantees to deliver a few of its operations throughout the pond: HPE shall be manufacturing liquid-cooled Cray EX supercomputers and Apollo methods in its Kutná Hora manufacturing unit within the Czech Republic (the place EURO IT4I shall be based mostly), and the corporate has additionally dedicated to establishing a Heart of Excellence in Europe that may present R&D providers and experience in assist of exascale readiness.

Three of the eight deliberate EuroHPC methods are but to be detailed:  the four-petaflops PetaSC system, hosted by Sofia Tech Park in Bulgaria; the ten-petaflops Deucalion system, hosted by the Minho Superior Computing Heart in Portugal; and the 200-petaflops MareNostrum 5 system, hosted by the Barcelona Supercomputing Heart in Spain. By strategy of elimination, MareNostrum 5 would be the final of the three pre-exascale methods to be totally introduced. “For the third machine, the tendering course of is in its very ultimate section now,” Jensen mentioned, “so there shall be bulletins on that within the coming weeks.”

LUMI is the most recent in a collection of wins for HPE’s Cray EX methods, which is able to function the idea for the three deliberate U.S. exascale methods (Aurora, El Capitan and Frontier) and Australia’s strongest supercomputer (simply announced this week). Conspicuously absent, nevertheless, are mentions of the European Processor Initiative (EPI) and its deliberate first-generation “Rhea” CPUs, ostensibly slated for 2021. Jensen maintains that the EPI continues to issue into the way forward for EuroHPC.

“The ambition is that after we get to the exascale methods – which is what we shall be discussing within the coming years – … that EPI will ship a processor that we are able to base at the least one of many upcoming exascale methods on,” he mentioned.

Header image: idea artwork of the LUMI supercomputer. Picture courtesy of CSC. 

Source link

Related posts

Los Angeles Lakers vs Denver Nuggets Game 3 Predictions, Odds & Picks


Editor’s Picks: September 14th 2020 – Office politics, property rights in poor countries and Mexico’s unsellable presidential jet | Podcasts


Psst, This is The place to Store Classic Furnishings in Singapore


Leave a Comment