IBM is unveiling a new system-on-chip designed to run and train deep learning models faster and more efficiently than a general-purpose CPU called the Artificial Intelligence Unit (AIU).
The chip, designed by the IBM Research AI Hardware Center, is dedicated to AI in the same way that diesel fuel is dedicated for diesel engines. The company says it has run deep learning models on CPUs and GPUs for the last decade, but what was really needed was an all-purpose chip optimized for the types of matrix and vector multiplication operations used for deep learning.
According to an IBM Research blog, the company has spent the last five years developing a design for a chip customized for the statistics of modern AI.
The company says there are two main paths to get there: a chip that isn’t as precise as a CPU that allows a drop from 32-bit floating point arithmetic to bit-formats holding a quarter as much information.
“This simplified format dramatically cuts the amount of number crunching needed to train and run an AI model, without sacrificing accuracy,” IBM researchers say.
These formats also reduce the drag on speed that comes from moving data to and from memory. The IBM AIU uses a range of smaller bit formats, including both floating point and integer representations, to make running an AI model less memory intensive.
IBM also says an AI chip should be laid out to streamline AI workflows with a simpler layout than a multi-purpose CPU. The chip has also been designed to send data directly from one compute engine to the next, resulting in “enormous energy savings,” the company says.
According to IBM, the AIU is an application-specific integrated circuit (ASIC), designed for deep learning and any deep-learning task, such as processing spoken language or words and images on a screen.
“Our complete system-on-chip features 32 processing cores and contains 23 billion transistors — roughly the same number packed into our z16 chip,” AI researchers write in a blog. “The IBM AIU is also designed to be as easy-to-use as a graphics card. It can be plugged into any computer or server with a PCIe slot.”
IBM says the AIU is the scaled version of an AI accelerator built into the company’s Telum chip, with the 2 cores in the AIU closely resembling the AI core embedded in the Telum chip that powers IBM’s z16 system.
“The AI cores built into Telum, and now, our first specialized AI chip, are the product of the AI Hardware Center’s aggressive roadmap to grow IBM’s AI-computing firepower,” IBM AI researchers say. “Because of the time and expense involved in training and running deep-learning models, we’ve barely scratched the surface of what AI can deliver, especially for enterprise.”
If you enjoyed this article and want to receive more valuable industry content like this, click here to sign up for our digital newsletters!
Leave a Reply