The French Atos formalized Monday at the end of the day the final delivery of Leonardo, a supercomputer with phenomenal power, which took its bases in Italy.
Atos announces that it has delivered the main part of the Leonardo system to the heart of the Bologna technology park, where it is now hosted, but also managed by the inter-university consortium Cineca, the largest Italian computing center. The latter comes, with Atos, to support the imperative mission of sovereignty of the European Union, within the framework of the fight against medical and environmental emergencies. Leonardo is certified 4e most powerful supercomputer in the world, according to the TOP500.
The compromise between incredible power (250 petaflops) and lower power consumption
It will have required more than 10,000 fiber optic cables and some 156 km of cables for its data interconnection network (155 equipment racks were delivered in total), but Leonardo will be able to do its job. Based on Atos’ BullSequana XH2000, the supercomputer offers the compromise of extremely high throughput and controlled energy consumption. On this last point, we somewhat regret the lack of data projection.
The second most powerful supercomputer in Europe even though its installation only started a few months ago, the AI-powered Leonardo will have a phenomenal power of 250 petaflops certified after the Linpack high performance test. Once it is complete, it will therefore have the capacity to carry out 250 million billion floating point operations per second.
This is more or less 10 times more than the previous system hosted by Cineca. On the storage power side, Leonardo will be equipped with more than 100 petabytes. Delayed due to the pandemic, access to the data center was finally quickly provided, at the end of July, to the European teams of Atos and Cineca.
To better understand, manage and anticipate extreme natural phenomena
Leonardo will use its capabilities to mitigate and manage risks due to so-called extreme environmental situations, such as volcanic eruptions, earthquakes, tsunamis, flash floods or other natural events, without forgetting the fight against pandemic and epidemic situations. Modeling these phenomena indeed requires high performance simulations, artificial intelligence, data analysis and visualization.
From a more technical perspective, the system is built on Atos BullSequana XH2000 nodes with direct liquid cooling (DLC). Each features four NVIDIA A100 Tensor Core GPUs and a single third-generation Intel Xeon Scalable evolution processor. To meet the demands of high-performance computing, it relies on Micron’s DDR5 DRAM technology, which provides the necessary bandwidth and system performance.
The system will use the NVIDIA Quantum 200 Gbit/s networking platform, with compute acceleration engines providing particularly low latency and high data throughput. It is also equipped with approximately 3,500 Intel Xeon processors and 14,000 NVIDIA A100 GPUs, offering a performance of 10 exaflops in reduced precision, ideal for AI applications. Finally, the data-centric partition is based on a BullSequana X2140 three-node CPU blade. It is equipped with two fourth-generation Intel Xeon Scalable (formerly “Sapphire Rapids”) processors, which each include 56 cores.
9