Harvard’s Kempner Institute expands academic computing cluster, adding nearly 400 GPUs | News Harvard crimson

Harvard’s Kempner Institute for the Study of Natural and Artificial Intelligence bought nearly 400 advanced graphics processing units last month to boost its computing cluster, specifically for training artificial intelligence models.

GPUs are computing units capable of central processing, memory and networking. With the addition of NVIDIA H100-80 GB graphics units, the Kempner Institute cluster has become one of the largest academic computing clusters in the world. Computing clusters consist of a collection of computers that work together to perform more computationally intensive tasks efficiently.

The Kempner Institute was launched in December 2021 with a mission to understand the basis of intelligence in natural and artificial systems. In the spring of 2022, the institute established its first computing cluster with 144 A100-40 GB GPUs.

This cluster includes a diverse range of hardware and software technologies that specialize in advanced machine learning and networking capabilities. This cluster enables large-scale experimentation and research in natural and artificial intelligence.

Elise Porter, CEO of Kempner We have a brilliant community of students and scientists pursuing ambitious machine learning projects and asking big, important questions, but some of these projects are only possible if we have access to the level of technology available in industry. . The institute wrote in an email statement.

The H100 model offers three times the training and inference speed of the original A100 model and is networked using Infiniband, an Internet computing architecture. The new addition uses a transfer rate of 1600 gigabytes per second, which trains language models faster and more efficiently.

This development is specifically designed to train generative AI agents, with increased computing power enabling faster development and testing of new models.

“Generative artificial intelligence is arguably one of the most important technological innovations of our generation, and these models are built on large clusters of highly networked superfast processors,” Porter wrote.

He added that building an academic cluster at scale, with GPUs this powerful, will enable scientists at Kempner to pursue cutting-edge machine learning research here at Harvard.

The processing units are located at the Massachusetts High Performance Computing Center, an intercollegiate computing center in western Massachusetts.

Camilla J. Staff writer Martinez can be reached at camilla.martinez@thecrimson.com. Follow him on X @camillajinm.

You can meet Tiffany A. Reach staff writer Mezitis at tiffani.mezitis@thecrimson.com.