In a groundbreaking development, scientists at Intel have successfully constructed the world's largest neuromorphic computer, named "Hala Point." This cutting-edge machine, designed to mimic the human brain, aims to support future artificial intelligence (AI) research and represents a significant leap in computational capabilities.

According to Intel representatives, Hala Point has demonstrated outstanding performance, surpassing conventional computing systems that rely on central processing units (CPUs) and graphics processing units (GPUs). The neuromorphic computer can execute AI workloads an impressive 50 times faster while utilizing 100 times less energy than its traditional counterparts.

The noteworthy figures are based on recent findings uploaded to the preprint server IEEE Explore on March 18. However, it is important to note that these results have not yet undergone peer review.

Initially, Hala Point will be deployed at Sandia National Laboratories in New Mexico, where teams of scientists will utilize the computer's immense capabilities to tackle critical problems in device physics, computing architecture, and computer science.

This state-of-the-art system is powered by 1,152 of Intel's innovative Loihi 2 processors, specially designed for neuromorphic research. Hala Point consists of a staggering 1.15 billion artificial neurons and an astounding 128 billion artificial synapses distributed across 140,544 processing cores. This enables the computer to perform a mind-boggling 20 quadrillion operations per second, or 20 petaops.

Comparing neuromorphic computers to supercomputers proves challenging due to their distinct data processing methods. Trinity, presently ranked as the 38th most potent supercomputer globally, boasts approximately 20 petaFLOPS of power. In contrast, the world's most powerful supercomputer, Frontier, operates at an impressive performance level of 1.2 exaFLOPS, equivalent to 1,194 petaFLOPS.

One key differentiating factor of neuromorphic computing is its unique architecture, as explained by Prasanna Date, a computer scientist with the Oak Ridge National Laboratory. Unlike classical computing, which involves binary bits flowing into hardware for sequential calculations, neuromorphic computing utilizes neural networks built within the machine. These networks process information in parallel, mimicking the human brain's functionality.

Hala Point and the Loihi 2 processors employ spiking neural networks (SNNs), where information is transmitted through discrete electrical signals known as "spike inputs." These networks are akin to the neural connections in the brain, allowing for parallel processing and measuring spike outputs following calculations. Additionally, the chips integrate memory and computing power, eliminating the bottleneck created by separate components in conventional computers.

Early results also indicate Hala Point's exceptional energy efficiency for AI workloads, with a reading of 15 trillion operations per watt (TOPS/W). This far outperforms most traditional neural processing units (NPUs) and other AI systems that typically achieve less than 10 TOPS/W, showcasing the significant strides the neuromorphic computer has made.

While neuromorphic computing is still an emerging field, with few machines like Hala Point currently deployed, researchers from the International Centre for Neuromorphic Systems (ICNS) at Western Sydney University in Australia have announced plans to deploy a similar computer named "DeepSouth" in December 2023. Their system will emulate vast networks of spiking neurons, achieving an impressive rate of 228 trillion synaptic operations per second, equivalent to the processing power of the human brain.

Intel representatives emphasize that Hala Point is a crucial starting point and a research prototype that will lay the foundation for future commercially viable neuromorphic systems. These systems might revolutionize AI deployments by enabling large language models (LLMs) like ChatGPT to learn continuously from new data, thereby reducing the training burden associated with current AI models.