Getting to know IBM's artificial intelligence unit

AI models are growing exponentially, but the hardware to train these giants and run them on servers in the cloud or on peripheral devices such as smartphones and sensors is not evolving as fast.

IBM Research's Artificial Intelligence Hardware Center decided to create a specialized computer chip for artificial intelligence, the Artificial Intelligence Unit, or AIU. It is the first complete system-on-a-chip designed to run and train deep-learning models faster and more efficiently than general-purpose CPUs.

Standard chips, known as CPUs, or central processing units, were developed before the deep learning revolution, a form of machine learning that makes predictions based on statistical patterns in large data sets. The flexibility and high accuracy of CPUs are good for general-purpose software applications, but not when it comes to learning and running deep-learning models that require massively parallel AI operations.

The IBM AIU, in return , is a so-called application-specific integrated circuit (ASIC). It is designed for deep learning and can be programmed to perform any type of deep learning task, whether it is processing spoken language or words and images on a screen. The BM AIU is also designed to be as easy to use as a video card. It can be connected to any computer or server with a PCIe slot.

In the meantime, here's a reminder of an article in Nature about AlphaTensor from DeepMind. It is the first artificial intelligence (AI) system to discover new, efficient, and provably correct algorithms for fundamental problems such as matrix multiplication. The AlphaTensor system is based on AlphaZero, an agent that has shown superhuman performance in board games such as chess, go, and shogi.