IBM introduces NorthPole chip to accelerate AI-based image recognition tasks
A team of scientists and engineers from IBM Research has developed a specialized NorthPole chip that can run AI-based image recognition applications 22 times faster than chips on the market.
What is known
According to experts, the use of edge systems, where data is physically closer to AI applications, can speed up their work.
At NorthPole, engineers have merged the processing module and the data it uses. The chip’s design is inspired by the human brain, researchers say.
To perform tasks, NorthPole uses a two-dimensional array of memory blocks and processors. The architecture allows for efficient data exchange between them and fast information processing.
In testing, NorthPole outperformed other chips, including NVIDIA GPUs, by a factor of 22 on typical image recognition tasks.
However, engineers admit that the accelerator is limited to highly specialized AI operations. IBM plans to test clusters of multiple NorthPole chips to expand its capabilities.
