Intel Unveils World’s Largest Neuromorphic System

Intel has unveiled the world’s largest neuromorphic computer, Hala Point, a device designed to mimic the workings of the human brain.

According to the developers, the system runs more complex AI models compared to conventional ECMs. The project is based on the Pohoiki Springs research cluster.

While traditional computing machines use a processor to perform operations and store data in separate memory, the neuromorphic device employs “artificial neurons.”

This architecture provides higher energy efficiency. Intel representatives claim that when solving optimization tasks, Hala Point consumes 100 times less energy compared to ordinary machines.

The device has the potential to unlock new algorithms for training and deploying neural networks. AI models will be able to use neural networks to process information similarly to a real brain.

Hala Point contains 1.15 billion “artificial neurons” across 1152 Loihi 2 chips and performs up to 380 trillion synaptic operations per second.

Mike Davies, Director of Intel Labs’ Neuromorphic Computing Lab, noted that despite its high power, the computer occupies only six racks in a server chassis.

Davies suggested the possibility of larger machines:

“There are no particular technical or engineering issues that would make us stop at this level.”

A distinctive feature of Hala Point is its ability to achieve high computational efficiency, exceeding 15 TOPS/W when deploying deep neural networks.

According to the developers, this level of power competes with and even surpasses architectures based on graphics and central processors.

Hala Point will not enter mass production. It is a research prototype that Intel has deployed at the Sandia National Laboratories, part of the National Nuclear Security Administration under the U.S. Department of Energy.

Back in April, Intel introduced the next-generation AI chip, Gaudi 3.