Intel Introduces Hala Point: The Next Generation of Neuromorphic Computing Systems

Intel unveiled its latest neuromorphic computer system, Hala Point, on Wednesday. Utilizing 1,152 Loihi 2 processors, it is designed to advance research on brain-inspired artificial intelligence (AI) and promote more sustainable AI applications.

Although announced today, Hala Point is a research prototype and not commercially available. Intel has deployed this neuromorphic system to Sandia National Laboratories, part of the U.S. Department of Energy's National Nuclear Security Administration (NNSA). This collaboration has been in place since 2021, aimed at further exploring neuromorphic computing in AI.

“The computing cost of today’s AI models is increasing unsustainably,” said Mike Davies, director of Intel Labs’ Neuromorphic Computing Lab. “We developed Hala Point to represent a new approach that combines the efficiency of deep learning with innovative, brain-inspired learning and optimization capabilities.”

Hala Point can support up to 30 quadrillion operations per second (30 petaops) and boasts an efficiency of over 15 trillion 8-bit operations per second per watt when executing conventional deep neural networks.

The system's architecture includes thousands of Loihi 2 processors, capable of supporting 1.15 billion neurons and 128 billion synapses across over 140,000 neuromorphic processing cores. Additionally, it features more than 2,300 embedded x86 processors and delivers significant memory bandwidth: 16 petabytes per second for memory, 11 PB/s for inter-core communication, and 5.5 terabytes per second for inter-chip communication.

Hala Point represents a significant evolution from Intel's first large-scale research system, Pohoiki Springs, offering a tenfold increase in neuron capacity and twelvefold performance enhancement.

“Applied to bio-inspired spiking neural network models, Hala Point can execute 1.15 billion neurons at speeds up to 20 times faster than a human brain and 200 times faster at lower capacities,” Davies noted. “While it’s not aimed at neuroscience modeling, its neuron capacity is roughly comparable to that of an owl brain or the cortex of a capuchin monkey.”

The advancements from Pohoiki Springs, alongside improvements in Loihi 2 architecture, position Hala Point to provide neuromorphic performance gains for mainstream deep learning models, particularly those requiring real-time processing like video, speech, and wireless communications.

Though unavailable to the public, Hala Point is believed to facilitate large-scale brain-based computing for Sandia Labs and NNSA research teams, addressing substantial challenges in physics, chemistry, and environmental science.

“Hala Point can solve optimization problems using 100 times less energy and at speeds up to 50 times faster than traditional CPU and GPU architectures,” Davies explained. “This exciting research realm leverages brain-inspired algorithms that differ significantly from those developed for conventional processing. Potential applications include logistics, vehicle fleet routing, railway scheduling, and smart city infrastructure management.”

Intel has not disclosed the cost of Hala Point, but access to smaller-scale systems is available to members of the Intel Neuromorphic Research Community through a free cloud platform open to academia, government, and corporate entities.