Tierney - stock.adobe.com

Inside India’s supercomputing journey

India is looking to shore up its supercomputing capabilities, but more needs to be done to realise its ambition of becoming a world leader in the field

What began as a mission to be self-dependent in supercomputing with the advent of Centre for Development of Advanced Computing (C-DAC) in India in the late ’80s has turned into a journey of new supercomputing milestones every year for India.

India-made supercomputers have been making it to top global lists, from the Param 8000 that debuted in 1991 to recent wonders like Param Siddhi from C-DAC and Param Pravega at the Indian Institute of Science.

But while Indian supercomputing feats have come a long way, the latest ones mark a new twist, with a new generation of artificial intelligence (AI) supercomputers that have been making their mark on the global stage.

In the 61st edition of the Top 500 global supercomputing list, India’s home-grown AI supercomputer, Airawat, which was installed at C-DAC in Pune under the Indian government’s national AI programme, was ranked 75th in the world.

At the time of the announcement, Airawat was integrated with Param Siddhi to deliver a total peak compute of 410 AI petaflops. The Ministry of Electronics and Information Technology expects Airawat to deliver 1,000 AI petaflops to cater to growing computational demands.

In the private sector, there is Yotta Infrastructure’s Shakti Cloud, touted as India’s largest and fastest supercomputer for AI and high-performance computing (HPC), as well as NeevCloud, which offers supercomputing services powered by 2,000 Nvidia H100 graphics processing units (GPUs).

Narendra Sen, NeevCloud’s founder and CEO, said the company aims to democratise access to AI and supercomputing for enterprises and startups, claiming to provide the world’s lowest prices for cloud GPUs at $1.69 per hour and reducing costs by up to 50%.

Read more about IT in India

  • Almost two-thirds of organisations in India said their responsible AI practices and policies were mature or they had taken steps towards responsible AI adoption, according to a Nasscom study.
  • India’s datacentre market is expected to surpass $10bn by 2028, but realising its full potential will require innovations in sustainability and improvements in connectivity.
  • India's progress in space technology will yield benefits not only for its burgeoning space industry but also propel IoT and telephony applications
  • Indian organisations are speeding up deployments of AI across multiple sectors, but legacy systems, siloed data and a shortage of AI-specific talent will stand in the way of greater adoption.

V Krishna Nandivada, head of the computer science and engineering department at the Indian Institute of Technology in Madras, noted that while there has been strong interest in HPC and supercomputing in India, the country does not have leadership in the field yet.

“But we definitely have the potential to be leaders on the global map,” he said. “With a large workforce, probably the largest in the world, as well as the application potential in a large and diverse country, we are uniquely placed to impact all dimensions of HPC research and development.”

With AI being the flavour of the season, the footprint of GPUs in supercomputers is hard to miss. The Frontier system that topped the supercomputer list has 8,699,904 combined CPU and GPU cores. In second place is the Aurora system, which houses Intel Data Center GPU Max Series accelerators, followed by Eagle, which packs Nvidia H100 accelerators.

Indranil Bandyopadhyay, principal analyst at Forrester, said advancements in AI-centric hardware will speed up model training, enhance the quality of algorithms and improve energy efficiency.

“Hardware is getting cheaper, smarter and better, and while it may not be specific to AI, this fillip will give a boost to AI,” he said. “Let’s not forget that AI was present 20 years back, but we now have the hardware to propel it.”

Homegrown machines

As a technology powerhouse, India is no stranger to building supercomputers. C-DAC, for one, has been producing Param supercomputers with hardware components sourced from local companies, along with a full software stack.

“We have efforts towards RISC-V based processors, and recently the government is encouraging companies to set up plants in India,” said Rupesh Nasre, head of the National Supercomputing Mission Nodal Centre at the Indian Institute of Technology in Madras. “With these efforts, it’s on the horizon to have close to 100% homegrown supercomputers, but these efforts need to be accelerated and nurtured through an HPC ecosystem which we currently lack.”

The scope of indigenisation is huge. Currently, the US, Europe and, more recently, China, are leading the world in HPC and supercomputing, “but considering the high-pace growth in this space, we have a great opportunity to lead and take the world along”, said Nandivada.

Sireesha Rodda, professor and head of the department of computer science and engineering at the Gandhi Institute of Technology and Management’s (Gitam) school of technology, said the first phase of India’s national supercomputing efforts involved deploying supercomputers with 60% Indian components.

The second phase will involve supercomputers with India-designed processors, and the third and fourth phases will see the development of fully indigenous supercomputers reaching the performance of 45 petaflops, she added.

But Chiranjib Sur, assistant professor at the Indian Institute of Technology in Guwahati’s data science and AI school, noted that indigenisation can happen only when India shores up its hardware manufacturing capabilities.

“That means everything starting from the fabrication of circuits, and not just assembling multiple parts and packaging them in a box. Today, the infrastructure is acquired at a very high cost compared to the US,” said Sur, noting that India’s basic customs duty of 20% is also “very high” for cloud providers that need an enormous amount of hardware to run supercomputing workloads.

Diverse workloads

AI supercomputers can be useful in many areas where speed, modelling and compute work are paramount, particularly in simulation environments such as digital twins.

“We run a lot of simulations when we make prototypes like a wind turbine,” said Kameswara Sridhar V from the department of mechanical engineering at Gitam’s school of technology. “We develop mathematical models which give us quicker responses and we can also test new scenarios.”

In healthcare, Rodda’s team has worked on the diagnosis of dental anomalies and diseases using deep learning techniques. The work was done on G-Cluster, a HPC cluster at Gitam that was rated as one of India’s top 50 HPC facilities.

Sur noted that the biggest application of supercomputing is the e-ticketing platform from Indian Railways that sells 130,000 tickets a day, or about 480 million tickets per year. The system was developed by Tata Consultancy Services, one of India’s big four IT service providers.

Nasre noted that India’s efforts to bolster its supercomputing capabilities are often isolated and rely heavily on government funding. “Government, industry and academia need to work together on a grand plan with milestones towards a self-sustained model of indigenous development,” he said. “The model also needs to permit and account for failures along the way.”

NeevCloud’s Sen pointed to another critical hurdle – the availability of advanced technologies such as Nvidia GPUs, essential for pushing the boundaries of HPC.

“If India has to become a sustainable economy and developed country by 2047, it will have to be leading in research and development in HPC software systems, applications and hardware,” said Nandivada.

Read more on Clustering for high availability and HPC

Data Center
Data Management