Nvidia to build 'world's fastest artificial intelligence supercomputer'

• The comparison
• Tutorials
• Nvidia to build 'world's fastest artificial intelligence supercomputer'

Nvidia and Cineca, an Italian inter-university consortium and large supercomputing center, have announced their intention to build "the world's fastest AI supercomputer." The upcoming Leonardo system will use nearly 14.000 Nvidia A100 GPUs for a variety of high-performance computing tasks. The maximum performance of the system should reach 10 FP16 ExaFLOPS. The supercomputer will be based on Atos's BullSequana XH2000 supercomputer nodes, each with an unknown Intel Xeon processor, four Nvidia A100 GPUs, and a Mellanox HDR 200 Gb/s InfiniBand card for connectivity. The slides are water cooled and there are 32 in each HPC cabinet.

(Image credit: Atos) The BullSequana XH2000 architecture is very flexible so it can accommodate any processor and GPU and for that purpose we can only guess which Intel Xeon processor will be used for the Leonardo.

(Image credit: Atos) Scientists at Italian universities plan to use Leonardo for drug discovery, space exploration and research, and weather modeling. Traditionally, these applications have relied on high-performance simulation and data analysis workloads that require FP64 precision. But Nvidia says that many HPC tasks today rely on powerful AI and machine learning, and for such workloads, FP16's precision is sufficient. Naturally, a large number of GPUs can also perform high-resolution displays. Nvidia's A100 GPU was built primarily for computing, so it supports all kinds of precision, including supercomputadora '' FP64 y la AI'' FP16. 14.000 Nvidia A100 GPUs can achieve performance up to 8.736 FP16 ExaFLOPS (624 TFLOPS per GPU with structural parsimony enabled × 14.000). Meanwhile, the same number of GPUs can deliver 135,800 FP64 TFLOPS, which is slightly lower than Summit's 148,600 TFLOPS FP64. Nvidia believes that AI and ML are crucial for today's supercomputer, so the company prefers to cite FP16's peak performance with structural parsimony enabled, in the case of the Leonardo supercomputer powered by its A100 GPUs. “With the advent of AI, we now have a new metric to measure supercomputers. As a result, the performance of our supercomputers has skyrocketed as their computing power has increased exponentially with the introduction of AI,” Ian Buck, vice president and general manager of Accelerated Computing at TechRadar Pro, told TechRadar. Nvidia. “Modern supercomputers today have to be artificial intelligence supercomputers to be an essential tool for science. Nvidia opens a new trend by combining HPC and AI. Only AI supercomputers can deliver 10 ExaFLOPS of AI performance with nearly 14.000 GPUs based on the NVIDIA Ampere architecture. Sources: Nvidia Press Release, Nvidia Blog Article