Recently, chip giant Intel has released a large-scale neuromorphic system called Hala Point, which was initially deployed at the Sandia National Laboratory in the United States to support future research in brain like AI and address current challenges in efficiency and sustainability.
It is reported that Hala Point was developed through architecture improvements based on Intel's first generation large-scale research system Pohoiki Springs, with a neuron capacity increase of more than 10 times and performance improvement of 12 times.
As the first large-scale neural mimetic system to demonstrate advanced computing efficiency in mainstream AI work, Hala Point supports up to 20 trillion operations per second (20 petaops) when executing traditional deep neural networks, with an efficiency exceeding 15 trillion 8-bit operations per watt per second (TOPS/W).
Intel stated that its unique features can enable real-time continuous learning for future AI applications, such as solving scientific and engineering problems, logistics, smart city infrastructure management, big language models, and artificial intelligence agents.
According to Intel's announcement, Hala Point is equipped with 1152 Intel Loihi 2 processors based on the Intel 4 process, supporting up to 1.15 billion neurons and 128 billion synapses. It can process over 380 trillion 8-bit synapses and over 240 trillion neuron operations per second; Its maximum power consumption is 2600 watts, and it also includes over 2300 embedded x86 processors for auxiliary computing.
When applied to biomimetic pulse neural network models, Hala Point can run its 1.15 billion neurons at a real-time speed 20 times faster than the human brain, with a maximum speed of up to 200 times faster than the human brain when the number of running neurons is low. Although Hala Point is not used for neuroscience modeling, its neuronal capacity is roughly equivalent to the brain of an owl or the cerebral cortex of a macaque. Intel also pointed out that this Loihi based system can perform AI inference and solve optimization problems 50 times faster than traditional CPU and GPU architectures.
Behind it lies Intel's long-term research on neural mimicry. In short, neural mimetic computing comes from learning the brain's computational processes, where the brain's neural network transmits information through pulses, and the collaborative and competitive interactions between multiple regions in the brain and its environment generate intelligent behavior. And neural mimicry, as a computational method, can better simulate the structure of human brain neurons and improve intelligence.
Since 2015, Intel has been conducting research on neuromorphic computing. In 2017, Intel launched its first self-learning neuromorphic chip, Loihi; By 2019, Intel had launched a Pohoiki Beach system with 64 Loihi chips, and by 2020, Intel's Pohoiki Springs had 768 Loihi chips and 100 million neurons. It is reported that Intel has chosen the talking Xuanfeng parrot for research.
In 2021, the Loihi series was upgraded, and Intel released the second-generation neuromorphic chip Loihi 2. Nowadays, Intel continues to iterate based on Loihi 2 and Pohoiki Springs. In the AI computing path, competition among chip giants is further intensifying, and research on brain like technologies and interdisciplinary research is also continuing to develop in controversy.
Mike Davies, Director of the Neuromimetic Computing Laboratory at Intel Labs, said, "The computing power cost of AI models is currently increasing at an unsustainable rate, and the industry needs new methods for scale expansion. To this end, we have developed Hala Point, which combines the efficiency of deep learning with novel brain like learning and optimization capabilities. We hope that through research on Hala Point, we can make progress and breakthroughs in the efficiency and adaptability of large-scale AI technology."
Researchers at Sandia National Laboratory plan to use Hala Point for advanced brain scale computing research, focusing on solving scientific computing problems in device physics, computer architecture, and computer and information science. Craig Vineyard, head of the Hala Point team at Sandia National Laboratory, also stated, "Collaborating with Hala Point has enhanced our Sandia team's ability to solve computational and scientific modeling problems. Using systems of this scale for research will enable us to keep up with the development of AI in areas such as business, defense, and basic science."
Intel also revealed that the next step is to deliver the Hala Point system to the Sandia National Laboratory, which means the first deployment of Intel's shared large-scale neuromorphic research system. The following development will enable neuromorphic computing applications to overcome the power and delay limitations of real-time deployment of AI capabilities in the real world.