Intel brings Hala Point to the world: it is a prototype of a so-called neuromorphic system that works more like the human brain. Intel hopes Hala Point will pave the way for efficient and affordable AI.

Intel shows Hala Point: a neuromorphic computing system based on the Loihi2 processor. Hala Point, like its predecessor Pohoiki Springs, must function more like the human brain and therefore has its own unique architecture. The system is based on (digital) neurons and Hala Point has ten times more than Pohoiki Springs. Memory and computing power are linked together in the solution, just like the brain.

Hala Point is an oversized server with 1,152 Loihi 2 processors housed in a 6U chassis. The system has 1.15 billion neurons and 128 billion synapses distributed across 140,544 neuromorphic processor cores. The system consumes 2.6 kW of power and has 2,304 x86 cores on board to support traditional computing.

The solution delivers 15 TOPS/W and is ideal for supporting deep learning networks. According to Intel, the neuromorphic approach is both more powerful and efficient than more traditional GPU-based AI hardware. Intel hopes Hala Point will make training large models like GPT-4 easier.

Challenger for Nvidia

The solution is a direct competitor to the ecosystem created by Nvidia and seems to have an advantage on paper. In practice, however, Nvidia can rely on an extensive and widely used package of software solutions. The question is to what extent AI developers are inclined to adopt a completely new architecture.

The Hala Point prototype is used in the American Sandia National Laboratories and is used there for computational problems in various physical domains. The computer system is very good at optimization problems compared to classical computer chips and has advantages that are also attributed to (not yet existing) quantum computers.