Centre’s AI push for Computational capacity

Centre’s AI push for Computational capacity

News: As part of an Artificial Intelligence Mission to develop its own ‘sovereign AI’, the Centre is looking to build computational capacity in the country and offer compute-as-a-service to India’s startups.

Background:
• The capacity building will be done both within the government and through a public-private partnership model, highlighting New Delhi’s intention to reap dividends of the impending AI boom which it envisions will be a crucial economic driver.

What is computing?
• Computing capacity, or compute, is among the most important elements of building a large AI system apart from algorithmic
innovation and datasets. It is also one of the most difficult elements to procure for smaller businesses looking to train and build such AI systems.

Scope and planning:
• In total, the country is looking to build a compute capacity of anywhere between 10,000 GPUs (graphic processing units) and 30,000 GPUs under the PPP model, and an additional 1,000-2,000 GPUs through the PSU Centre for Development of Advanced Computing (C-DAC).
• For context, according to a 2020 blog by Microsoft, the company had developed a supercomputer for OpenAI – the firm behind ChatGPT – which consisted of 10,000 GPUs among other things.
• A Cabinet approval would be needed to clear the proposal, with the government exploring various incentive structures for private companies to set up computing centres in the country.
• Efforts are being made to build Compute capacity within the C-DAC under the National Supercomputing Mission. They already have the Rudra and Param systems and we are planning to add 1,000-2,000 GPUs to them.
• Rudra is an indigenous server platform built by the C-DAC which has two expansion slots for graphic cards. Param Utkarsh is a high performance computing system setup at C-DAC which offers AI over machine learning and deep learning frameworks, compute and storage as a cloud service.

What is government’s planning?
• The government’s idea is to create a digital public infrastructure (DPI) out of the GPU assembly it sets up so that startups can utilise its computational capacity for a fraction of the cost, without needing to invest in GPUs which are often the biggest cost centre of such operations.