As tech giant Nvidia showcased its next-gen Graphic Processing Units (GPUs) and AI platforms at its annual GPU Technology Conference (GTC), the work of a research team from Indian Institute of Technology (IIT) - Madras found a mention at the event.

Nvidia supports the research work done by the Center for Computational Brain Research (CCBR) at IIT-Madras with computing hardware, computational systems, domain-specific libraries, curated deep-learning models, and other tools. This is critical for the centre to power a large-scale, multi-disciplinary effort to map more than 100 human brains at a cellular level. 

This partnership and the work of the centre was highlighted at GTC 2024 by Kimberly Powell, vice president of healthcare at NVIDIA, in her keynote speech.

“We have an amazing collaboration with the IIT Madras and their computational brain research centre... The team’s brain research platform called Neuro Voyager is imaging the brain at a cellular level with each brain image containing between 2-3 petabytes of data. We have worked with them to digitise the brain data and visualise it at any resolution right down to a half a micron,” Powell said. “They are also putting a chat front end on it. The chatbot can be used to query information from 10 years worth of brain research, and answer simple questions,” Powell added.

Also read: IIT Madras develops electric standing wheelchair

Mohanshankar S, professor of electrical engineering and head of CCBR told businessline that the team is working with unprecedented levels of dataset and also complex data, and thus right computing infrastructure is critical to expedite research and compete effectively. Most AI research is computing infrastructure-driven and academia partnerships with either industry or government will be essential going forward, he added.

 A joint note by IIT-Madras and Nvidia in November 2023 spoke about how the centre scans and maps brains at a rate of 250 GB per hour per scanner, and operating multiple scanners simultaneously they generate 2 TB per hour of high-resolution uncompressed images. “This is why the brain centre has turned to Nvidia, a leader in GPU-based HPC offerings, and has operationalized a cluster of DGX A100 systems to do the complete processing of 10 to 20 brains,” the note said. The note also said that the centre uses eight NVIDIA A100 Tensor Core GPUs per DGX node, and as a result, the data that requires a minimum of one hour to detect cells has been reduced to less than 10 minutes. IIT-M’s CCBR and its human brain project aims to develop a digital collection of human brains at a cellular resolution to help detailed exploration from individual cells to the entire brain structure. The insights from this project will help in areas like brain surgery, therapy, and understanding of neurological diseases.