In January, Italy’s Ministry of Schooling, College, and Analysis (MUIR), Nationwide Institute of Nuclear Physics, and World College of Complex Research unveiled Leonardo, a brand new supercomputer to be built at CINECA in a position to “exascale” computing for analysis and innovation. The companions for the venture, which used to be authorized via the Eu Joint Enterprise EuroHPC, a joint supercomputing collaboration between nationwide governments and the Eu Union, weren’t first of all published. However Nvidia nowadays showed that it’s going to provide Ampere-based graphics playing cards and Mellanox HDR 200GB/s Infiniband networking to Leonardo to ship as much as 10 exaflops of half-precision floating-point (FP16) efficiency.
Leonardo, which shall be funded via MUIR, is anticipated to take care of computing workloads involving drug discovery, house exploration, and climate modeling. Scientists shall be supplied get right of entry to to Leonardo to spot proteins that may be focused with explicit medicine and to are expecting excessive climate prerequisites, in addition to to research information from electromagnetic waves, gravitational waves, and neutrinos.
Leonardo shall be constructed from Atos Sequana nodes, each and every with 4 Nvidia Tensor Core graphics playing cards and a unmarried Intel processor of an unknown structure. (Atos is an Nvidia techniques spouse headquartered in France.) The Mellanox HDR InfiniBand Dragonfly+ connectivity will function in-network computing engines that permit low latency and prime throughput, whilst the Ampere graphics playing cards will be capable to accelerating over 1,800 recurrently used packages via as much as 70 occasions, together with Quantum Coffee for subject material science, SPECFEM3D for geoscience, and MILC for quantum physics. At the working device facet, Nvidia says Leonardo will run the similar CUDA device as CINECA’s current Nvidia-powered device.
Leonardo will sign up for a community of supercomputers — EuroHPC — deliberate for the Czech Republic, Luxembourg, and Slovenia. The Luxembourg-based MelaXina device, which can center of attention on monetary services and products, production, and well being care packages, will attach 800 Nvidia A100 graphics playing cards on HDR 200Gbps InfiniBand hyperlinks for as much as 500 petaflops of efficiency. As for the brand new Vega supercomputer on the Institute of Data Science in Maribor, Slovenia, it’s going to come with 240 A100 graphics playing cards and 1,800 HDR 200Gbps InfiniBand endpoints. Finally, The IT4Innovations Nationwide Supercomputing Middle will host what’s anticipated to develop into probably the most robust supercomputer within the Czech Republic: an Apollo 6500-based device with 560 A100 graphics playing cards to ship just about 350 petaflops of efficiency for tutorial and business simulations, information analytics, and AI.
Leonardo, MelaXina, Vega, and the IT4Innovations system are the newest in a sequence of supercomputing wins for Nvidia. In October at its annual GPU Generation Convention, the corporate published it’s going to give a contribution and experience to what’s anticipated to be the U.Okay.’s quickest supercomputer, Cambridge-1. And in July, Nvidia introduced plans to construct what it claims would be the quickest supercomputer in academia via bettering the features of the College of Florida’s HiPerGator supercomputer with the corporate’s DGX SuperPod structure.
“The EuroHPC generation roadmap for exascale in Europe is opening doorways for fast expansion and innovation in HPC and AI,” Marc Hamilton, Nvidia VP of answers structure and engineering, mentioned. “We’re operating with CINECA and Atos to boost up clinical discovery throughout a extensive vary of utility domain names, offering a platform to herald the technology of exascale computing.”