India is among first four countries to order the NVIDIA system
Pune; October 6, 2020 . The Centre for Development of Advanced Computing (C-DAC) has announced that it will shortly commission India’s fastest High Performance Computing (HPC) Artificial Intelligence (AI) supercomputer to be names ‘PARAM Siddhi – AI’.
The machine will deliver 210 AI Petaflops (6.5 Petaflops Peak DP), and will consist of 42 blocks of 42 of the NVIDIA DGX A100 universal system for AI workloads, each of which can deliver 5 petaflops ( 5x 42= 210). Together they will reconstruct the architecture of the NVIDIA DGX SuperPOD -- which is currently number seven in the world’s ranking of Top 500 supercomputers.
The hardware will be entirely imported from US-based NVIDIA. Software for the HPC-AI engine, Software Frameworks, Cloud Platform will be developed by C-DAC.
NVIDIA CEO Jensen Huang announced at the GPU Technology conference, October 5, that DGX SuperPODs are now shipping and are expected to be installed in Korea, the U.K., Sweden and India before the end of the year.
Says Dr Hemant Darbari, Director General, C-DAC: "PARAM Siddhi – AI, the State-of-the-Art large-scale HPC-AI scalable infrastructure that will be established under National Supercomputing Mission (NSM) at Centre for Development of Advanced Computing (C-DAC) with support from Niti Aayog, Ministry of Electronics and IT, Department of Science and Technology, Government of India using NVIDIA Next Generation technology, C-DAC software stack and Cloud platform will play a pivotal role in developing a vibrant ecosystem for research and innovation in science and engineering. This infrastructure will accelerate experiments and outcomes for India specific grand challenge problems in Health Care, Education, Energy, Cyber Security, Space, Automotive and Agriculture. It will catalyze partnerships with the Academia, Industry, MSMEs and Start-ups."
We are featuring a video on the NVIDIA system in the Tech Video spot of our home page
IndiaTechOnline comment: Having bought multiple Bull/Atos supercomouting systems in the first phase of the National Supercomputing Mission, India seemingly has decided to continue taking the 'buy' rather than 'build' route to attain competence in high performance computing. The NVIDIA acquisition today dwarfs the 2 XC 40 petaflop systems acquired from Cray that currently make it to the global Top 500 rankings, that are capable of 2.5- 3.7 peta flops.
With little visibility of indigenous development delivering a platform capable of tens of petaflops, in the near future, those deciding things in the NSM seem to have gone for an import-fueled short cut to access this level of computational muscle. This may be pragmatic. But it leaves unaddressed the compulsion for India to 'own' the IP required to build her own supercomputers for hardnosed stategic reasons. The NSM started with a kitty of Rs 45 billion (Rs 4500 crore). But how much will be left for indigenous development by the time the 'buy' phase is over?