0
Intel and Sandia National Labs Roll Out 1.15B Neuron “Hala Point” Neuromorphic Research System

Intel and Sandia National Labs Roll Out 1.15B Neuron “Hala Point” Neuromorphic Research System

While neuromorphic computing currently remains under research, efforts in this area have grown steadily over the years, as have the capabilities of specialized chips designed for this research. Following these lines, this morning Intel and Sandia National Laboratories are celebrating the deployment of the HalaPoint Neuromorphic System, which both believe is the most capable system in the world. With a total of 1.15 billion neurons, HalaPoint is the largest deployment for Intel to date. Iron 2 neuromorphic chipWhich was first announced in late 2021.

The HalaPoint system includes 1152 Iron 2 processors, each capable of simulating a million neurons. As noted at the launch of the Lohi 2, these chips are actually very small – just 31mm.2 With 2.3 billion transistors per chip, because they are built on the Intel 4 process (besides Meteor Lake, one of the only Intel chips to do so). As a result, the complete system is correspondingly small, taking up only 6 rack units of space (or as Sandia likes to compare it to the size of a microwave), with a power consumption of 2.6 kW. Now that it's online, the Hala point is gone. The SpiNNaker system As the largest disclosed neuromorphic system, it offers only a small number of neurons recognized at less than 3% of the 100 kW British system.



A single iron 2 chip (31 mm2)

Hala Point Sandia will replace an older Intel neuromorphic system at Pohoiki Springs, based on Intel's first-generation Loihi chips. In comparison, the HALA Point offers ten times more neurons, and 12 times more efficiency overall.

Both neuromorphic systems have been acquired by Scindia to advance the national lab's research into neuromorphic computing, a computing model that behaves like a brain. The main idea (if you'll excuse the pun) is that, mimicking the wet stuff that wrote this article, neuromorphic chips can be used to solve problems that traditional processors can't solve today. , and that they can do it more effectively.

Sandia, for its part, has said it will use the system to look at large-scale neuromorphic computing, at scales beyond Pohaiki Springs. With the halo point offering an artificial neuron count nearly at the level of complexity of an owl's brain, the lab believes a large-scale system will eventually enable them to precisely harness the properties of neuromorphic computing. So that it can solve real problems in areas like equipment. Physics, computer architecture, computer science and informatics are moving beyond simple phenomena initially obtained on a small scale.

A new focus from the lab, which has in turn attracted Intel's attention, is the application of neuromorphic computing to AI inference. Since the neural networks behind the current wave of AI systems are trying to emulate the human brain itself, there is, in a sense, a clear degree of convergence with brain-mimicking neuromorphic chips, even if the algorithms do some important things. I am different. Still, with energy efficiency being one of the big advantages of neuromorphic computing, it's prompted Intel to look into the matter further — and even build a second, halo-point-sized system of its own. .

According to Intel, in their research at Hala Point, the system has achieved performance as high as 15 TOPS-per-Wat at 8-bit precision, though using 10:1 sparsity compared to current-generation commercial chips. Made more competitive. . As an added bonus to this performance, neuromorphic systems do not require extensive data processing and batching in advance, which is usually necessary to make efficient use of high-density ALU arrays in GPUs and GPU-like processors.

Perhaps the most interesting use case of all, however, is the ability to be able to use neuromorphic computing to augment neural networks with additional data on the fly. The idea behind this is to avoid retraining, as required by current LLMs, which is prohibitively expensive due to the need for extensive computing resources. In essence, it's taking another page from how the brain works, allowing for continuous learning and growing datasets.

But for now, at least, it remains a subject of academic study. Ultimately, Intel and Sandia want systems like Hala Point to lead to commercial system development – ​​and, presumably, on a larger scale. But to get there, researchers at Sandia and elsewhere will first need to use the current crop of systems to better refine their algorithms, as well as better understand how this computing power works. How to map large workloads in style to demonstrate their utility at scale scles.CP

About the Author

Leave a Reply