Images above show ( top to bottom): The GPU core of the NVIDIA DGX A 100 systdem; the basic unit or building block of the DGX A100 and a comparison of a conventional High Performance Computing Data Centre with one fuelled by DGX A 100 which according to NVIDIAS occupies 1/25th the space, consumes 1/20th the power and costs 1/10th the price of similar systems
November 18 2020: India has three systems in the Top500 ranking of the world’s supercomputers released this week:
Rank/ system and location/ cores/Rmax
63/PARAM Siddhi-AI - NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Mellanox HDR Infiniband, Atos; Center for Development of Advanced Computing (C-DAC)/41664/4619 TFlops/
78/Pratyush - Cray XC40, Xeon E5-2695v4 18C 2.1GHz, Aries interconnect , Cray/HPE Indian Institute of Tropical Meteorology/119,232/3763.9 TFlops
146/Mihir - Cray XC40, Xeon E5-2695v4 18C 2.1GHz, Aries interconnect , Cray/HPE National Centre for Medium Range Weather Forecasting/83,592/2808.7 TFlops
The fastest India-based platform –Param Siddhi-AI, is based on the acquisition of 42 blocks of NVIDIA DGX A100 universal system for AI workloads which was commissioned last month at the Centre for Development of Advanced Computing (CDAC) Pune with the Atos Codex AI Suite and HPC-AI engine and software developed indigenously.