A neuromorphic chip: a new alternative to GPUs for AI

Scientists at Sandia National Laboratories, a nuclear laboratory run by the US Department of Energy, have announced that Intel's Loihi 2 neuromorphic chip can produce differential equations in partial derivatives using the finite element method. Initially, tasks related to fluid dynamics and weather modeling were only solved on supercomputers.

Features of neuromorphic computers

Neuromorphic machines function in a special way: they mimic the architecture of the brain in a certain manner, as a result of which memory and computation are combined in the same elements. Previously, it was believed that such computers could only be used to accelerate neural networks and image recognition. However, neurobiologists have developed a new algorithm called NeuroFEM, which can be used to translate the finite element method into a spiking neural network. The system, which was launched on 32 Loihi 2 chips, showed excellent scalability and good accuracy.

At this stage, the energy advantage of the new approach does not differ significantly from conventional systems. Large-scale comparisons and analysis require large tasks. However, in the long term, neuromorphic systems can significantly reduce costs.

The largest neuromorphic system, Hala Point, has already been delivered to the nuclear laboratory, where it will be tested for AI tasks. If neuromorphic chips can be used for AI and scientific calculations in the future, they could become a good alternative to GPUs.

Hope you found this article helpful - what do you think? Like and subscribe to our blog to get more practical recommendations and the latest tech news from HostZealot.