Intel introduces the most extensive AI ‘neuromorphic computer’ to date, modeled after the human brain. The Hala Point neuromorphic computer, powered by over 1,000 new AI chips, delivers a performance 50 times faster than traditional computing systems. Intel’s scientists have constructed this revolutionary machine to propel future artificial intelligence (AI) research.
Dubbed “Hala Point,” the system achieves remarkable efficiency, performing AI tasks 50 times faster while consuming 100 times less energy compared to conventional central processing unit (CPU) and graphics processing unit (GPU) systems. These advancements, outlined in a study uploaded to the preprint server IEEE Explore on March 18, mark a significant leap in AI computing capabilities.
Initially deployed at Sandia National Laboratories in New Mexico, Hala Point will tackle challenges in device physics, computing architecture, and computer science. Powered by 1,152 Loihi 2 processors, Intel’s neuromorphic research chip, this massive system comprises 1.15 billion artificial neurons and 128 billion artificial synapses distributed across 140,544 processing cores.
Capable of executing 20 quadrillion operations per second (20 petaops), neuromorphic computers like Hala Point process data differently from supercomputers, making direct comparisons challenging. Neuromorphic computing, as explained by Prasanna Date, a computer scientist at Oak Ridge National Laboratory, employs spiking neural networks (SNNs) to simulate the brain’s neural processes.
Unlike traditional computing, which relies on binary bits, neuromorphic computing uses discrete electrical signals to stimulate SNNs, enabling parallel processing and reducing energy consumption. The integration of memory and computing power in neuromorphic chips further enhances efficiency by eliminating data travel bottlenecks.
Early assessments reveal Hala Point’s outstanding energy efficiency, achieving 15 trillion operations per watt (TOPS/W) for AI workloads, surpassing conventional neural processing units (NPUs) and other AI systems. While neuromorphic computing remains an evolving field, Hala Point represents a significant milestone, paving the way for future commercial applications.
In December 2023, researchers at the International Centre for Neuromorphic Systems (ICNS) announced plans to deploy a similar system named “DeepSouth,” capable of emulating vast networks of spiking neurons at 228 trillion synaptic operations per second. These advancements herald a new era in AI computing, promising continuous learning and enhanced efficiency for future applications like large language models (LLMs) such as ChatGPT.
