Nvidia will supply hardware for Leonardo,10 exaflop supercomputer
Nvidia today confirmed that it will supply Ampere-based graphics cards and Mellanox HDR 200GB/s Infiniband networking to Leonardo to deliver up to 10 exaflops of half-precision floating-point (FP16) performance.
Leonardo, which will be funded through MUIR, is expected to handle computing workloads involving drug discovery, space exploration, and weather modeling. Scientists will be provided access to Leonardo to identify proteins that can be targeted with specific drugs and to predict extreme weather conditions, as well as to analyze data from electromagnetic waves, gravitational waves, and neutrinos.
Leonardo will be built from Atos Sequana nodes, each with four Nvidia Tensor Core graphics cards and a single Intel processor of an unknown architecture. (Atos is an Nvidia systems partner headquartered in France.) The Mellanox HDR InfiniBand Dragonfly+ connectivity will feature in-network computing engines that enable low latency and high throughput, while the Ampere graphics cards will be capable of accelerating over 1,800 commonly used applications by up to 70 times, including Quantum Espresso for material science, SPECFEM3D for geoscience, and MILC for quantum physics. On the operating system side, Nvidia says Leonardo will run the same CUDA software as CINECA’s existing Nvidia-powered system.
Leonardo will join a network of supercomputers — EuroHPC — planned for the Czech Republic, Luxembourg, and Slovenia. The Luxembourg-based MelaXina system, which will focus on financial services, manufacturing, and health care applications, will connect 800 Nvidia A100 graphics cards on HDR 200Gbps InfiniBand links for up to 500 petaflops of performance. As for the new Vega supercomputer at the Institute of Information Science in Maribor, Slovenia, it will include 240 A100 graphics cards and 1,800 HDR 200Gbps InfiniBand endpoints. Lastly, The IT4Innovations National Supercomputing Center will host what’s expected to become the most powerful supercomputer in the Czech Republic: an Apollo 6500-based system with 560 A100 graphics cards to deliver nearly 350 petaflops of performance for academic and industrial simulations, data analytics, and AI.
Leonardo, MelaXina, Vega, and the IT4Innovations machine are the latest in a series of supercomputing wins for Nvidia. In October at its annual GPU Technology Conference, the company revealed it will contribute hardware and expertise to what is expected to be the U.K.’s fastest supercomputer, Cambridge-1. And in July, Nvidia announced plans to build what it claims will be the fastest supercomputer in academia by enhancing the capabilities of the University of Florida’s HiPerGator supercomputer with the company’s DGX SuperPod architecture.
“The EuroHPC technology roadmap for exascale in Europe is opening doors for rapid growth and innovation in HPC and AI,” Marc Hamilton, Nvidia VP of solutions architecture and engineering, said. “We’re working with CINECA and Atos to accelerate scientific discovery across a broad range of application domains, providing a platform to usher in the era of exascale computing.”