Sunday, February 16, 2025

Hala Point Intel Creates World’s Largest Neuromorphic System

To enable more sustainable AI, Intel creates the largest neuromorphic system in the world. The first 1.15 billion neurone neuromorphic system in the industry, Hala Point Intel, paves the way for more effective and scalable AI.

Intel Hala Point

Intel declared that it has constructed the biggest neuromorphic system in the world. This large-scale neuromorphic system, code-named Hala Point, was first installed at Sandia National Laboratories. It makes use of Intel’s Loihi 2 processor, supports research for future brain-inspired artificial intelligence (AI), and addresses issues with the sustainability and efficiency of current AI. Hala Point incorporates architectural enhancements to reach up to 12 times greater performance and more than 10 times more neurone capacity in Intel’s first-generation large-scale research system, Pohoiki Springs.

How It Works

On common AI workloads, Hala Point Intel is the first large-scale neuromorphic system to exhibit cutting-edge computing savings. According to characterization, it can execute typical deep neural networks with an efficiency of about 15 trillion 8-bit operations per second per watt (TOPS/W) and handle up to 20 quadrillion operations per second, or 20 petaops. This equals or surpasses the performance attained by designs based on central processing units (CPU) and graphics processing units (GPU).

Future real-time continuous learning for AI applications including logistics, smart city infrastructure management, AI agents, large language models (LLMs), and scientific and technical problem-solving may be made possible by Hala Point’s special characteristics.

How It Will Be Used

Sandia National Laboratories researchers intend to use Hala Point Intel for cutting-edge brain-scale computing studies. The group will concentrate on resolving scientific computing issues in informatics, computer science, device physics, and computer architecture.

Hala Point is a research prototype that will enhance the functionality of commercial systems in the future. According to Intel, these lessons will result in useful developments like LLMs’ capacity to continually learn from fresh data. These developments might greatly lessen the unsustainable training load associated with large-scale AI implementations.

Why It Is Important

The necessity for innovation at the lowest levels of hardware design has been brought to light by recent trends in scaling up deep learning models to trillions of parameters, which have shown difficult sustainability difficulties in AI. In order to reduce data transportation, neuromorphic computing a radically novel technique based on insights from neuroscience integrates memory and processing with extremely granular parallelism.

This month’s International Conference on Acoustics, Speech, and Signal Processing (ICASSP) results showed that Loihi 2 improved the efficiency, speed, and flexibility of new small-scale edge workloads by orders of magnitude.

With several enhancements over its predecessor, Pohoiki Springs, Hala Point Intel now introduces neuromorphic performance and efficiency increases to mainstream traditional deep learning models, particularly those handling real-time workloads like wireless communications, voice, and video. As demonstrated at this year’s Mobile World Congress, Ericsson Research, for instance, is using Loihi 2 to maximize the efficiency of telecom infrastructure.

Hala Point Intel
Image Credit To Intel

About Hala Point Intel

Hala Point Intel is based on Loihi 2 neuromorphic processors, which use brain-inspired computing concepts like integrated memory and computing, asynchronous, event-based spiking neural networks (SNNs), and sparse and continuously changing connections to achieve orders-of-magnitude improvements in performance and energy consumption. Overall power consumption is decreased since neurones interact directly with one another instead of using memory.

Hala Point uses an Intel 4 process node to bundle 1,152 Loihi 2 processors in a microwave-oven-sized data centre chassis with six rack units. With a maximum power consumption of 2,600 watts, the system can accommodate up to 1.15 billion neurones and 128 billion synapses spread over 140,544 neuromorphic computing units. More than 2,300 integrated x86 processors are also included for auxiliary calculations.

Loihi 2 processors
Image Credit To Intel

Hala Point Intel offers a total of 16 petabytes per second (PB/s) of memory bandwidth, 3.5 PB/s of inter-core communication bandwidth, and 5 terabytes per second (TB/s) of inter-chip communication bandwidth by integrating processor, memory, and communication channels in a massively parallelised fabric. More than 240 trillion neurone activities and 380 trillion 8-bit synapses may be processed by the system in a second.

When used with bio-inspired spiking neural network models, the system can operate at up to 200 times the speed of the human brain at lesser capacity and at its maximum capacity of 1.15 billion neurones 20 times quicker. Hala Point’s neurone capacity is about the same as that of an owl’s brain or a capuchin monkey’s cortex, despite the fact that it is not designed for neuroscience modelling.

Compared to traditional CPU and GPU architectures, Loihi-based systems can solve optimization issues and conduct AI inference with 100 times less energy and up to 50 times quicker speeds. Early findings on Hala Point demonstrate that the system can achieve deep neural network efficiencies as high as 15 TOPS/W2 by taking advantage of up to 10:1 sparse connectivity and event-driven activity.

This is done without the need for input data to be gathered in batches, a common GPU optimization that considerably slows down the processing of data that arrives in real-time, like camera video. Future neuromorphic LLMs with continuous learning capabilities, however still in the development stage, might save gigawatt-hours of energy by removing the need for retraining on ever-increasing datasets.

What’s Next

Intel intends to share a new family of large-scale neuromorphic research systems with its research partners, and the transfer of Hala Point Intel to Sandia National Labs is the first deployment of this family. Applications of neuromorphic computing will be able to overcome power and latency limitations that prevent the real-world, real-time deployment of AI capabilities with further research.

In addition to an ecosystem of over 200 Intel is working to push the limits of brain-inspired AI and advance this technology from research prototypes to industry-leading commercial products over the next several years. Members of the Neuromorphic Research Community (INRC) include top academic groups, government labs, research institutions, and businesses worldwide.

Drakshi
Drakshi
Since June 2023, Drakshi has been writing articles of Artificial Intelligence for govindhtech. She was a postgraduate in business administration. She was an enthusiast of Artificial Intelligence.
RELATED ARTICLES

Recent Posts

Popular Post

Govindhtech.com Would you like to receive notifications on latest updates? No Yes