India’s Supercomputing Muscle Being Beefed-Up With New Machines For Weather Forecasting, And Partnerships With NVIDIA For Reliance And Tatas

There is no change in India’s entries in the global biannual Top 500 supercomputer list released last week.

Airawat (AI Research Analytics and Knowledge Dissemination Platform), clocking 8.5 petaflops/sec at  the Centre for Development of Advanced Computing (C-DAC), Pune under the initiative of Ministry of Electronics and IT,  remains the fastest supercomputer in India.

The second fastest supercomputer in India is  also to be found at C-DAC: the Param Siddhi which attained a  top speed of 4.62  petaflops. It is a Bull Sequana machine acquired from Eviden, the supercomputing arm of France-based Atos, which is working with C-DAC to progressively shift the assembly of its supercomputers to India. It is ranked number 163 in the current Top 500.

The number 3 India-based supercomputer is  Pratyush, a  Cray XC40 supercomputer from HPE with max speed of 3.76 petaflops  housed in the Indian Institute of Tropical Meteorology, Pune and ranked number 201.

Global First

When it comes to the global first, the top slot remains with the Oak Ridge National Laboratory (ORNL) in Tennessee, USA, and its Frontier computer, HPE-Cary machine  with a computational top speed of 1.194 ExaFlop/s. It remains the only exascale machine in Top 500.

In June this year, Eviden received a contract from India’s Ministry of Earth Sciences to supply two  new supercomputers for weather modelling and climate research: an 8.3 petaflops machine for the  NCMRWF and  a 13 petaflops machine for IITM. Both supercomputers will be part of the Bull Sequana XH2000 range.

Reliance-Tata-NVIDIA Tie-Ups

Founder-CEO, Jensen Huang of the graphical and AI-driven computing leader, NVIDIA announced in September during a visit to India, that Reliance Industries and Tata Communications will build and operate state-of-the-art AI supercomputing data centres based on NVIDIA technology like the GH 200 Grace Hopper Superchip and the DGX Cloud.

They will utilise this for internal AI development and infrastructure-as-a-service for India’s AI researchers, companies and burgeoning AI startup ecosystem. (The Grace Hopper superchip is a combo of central processor and graphics processor for giant-scale AI applications)

The joint release with Reliance said the company will create AI applications and services for their 450 million Jio customers and provide energy-efficient AI infrastructure to scientists, developers and startups across India.

AI can help rural farmers interact via cell phones in their local language to get weather information and crop prices. It can help provide, at massive scale, expert diagnosis of medical symptoms and imaging scans where doctors may not be immediately available.

AI can also  better predict cyclonic storms using decades of atmospheric data, enabling those at risk to evacuate and find shelter.

Said the chairman of Reliance Jio Infocomm, Akash Ambani about the partnership with NVIDIA: "Together, we will develop a state-of-the-art AI cloud infrastructure that is secure, sustainable and deeply relevant across India, accelerating the nation's journey towards becoming an AI powerhouse.”

Tata Sons’ chairman N Chandrasekaran said: “Our partnership with NVIDIA will democratize access to AI infrastructure, accelerate build-out of AI solutions and enable upgradation of AI talent at scale.” 

Jensen Huang suggests: “AI could be built in India, used in India, and exported from India.”

Tech Note

Computer speeds are measured in floating-point operations per second or FLOPS. These are simple arithmetic operations, like addition or multiplication, involving a number containing a decimal, say 2.5.

A person can typically solve an operation such as addition with a pencil and paper in one second — that would rate as one FLOP.

An exascale computer can perform more than 1,000,000,000,000,000,000 FLOPS, or 1 exaflop that is one followed by 18 zeros. In comparison a petaflop or one thousand teraflops is represented by one followed by 15 zeros.

We are familiar today with gigahertz speeds or gigabytes of storage. To put it in perspective, a petaflop is 1 million gigaflops. All supercomputers today except the world’s number 1, the Frontier, are as yet petaflop machines.