IBM Unveils Its Artificial Intelligence Unit, an Accelerator SoC for Deep Learning Workloads
Packing 32 cores, each based on the single accelerator in the Telum processor, this system-on-chip promises big efficiency gains.
Computing grandee IBM has announced its first complete system-on-chip (SoC) designed to train and run deep learning models quicker than general-purpose CPUs: the IBM Artificial Intelligence Unit (AIU).
"We're running out of computing power. AI models are growing exponentially, but the hardware to train these behemoths and run them on servers in the cloud or on edge devices like smartphones and sensors hasn’t advanced as quickly," claims IBM of the need for its new part. "That's why the IBM Research AI Hardware Center decided to create a specialized computer chip for AI. We’re calling it an Artificial Intelligence Unit, or AIU."
The IBM AUI is built around the concept of "approximate computing," which drops from 32-bit floating-point to lower precision formats in order to reduce the computational grunt and amount of shuffling of data between processor and memory required to train and run a given model — without, the company claims, sacrificing accuracy.
"It’s designed for deep learning and can be programmed to run any type of deep learning task, whether that’s processing spoken language or words and images on a screen," IBM claims of the AIU. "Our complete system-on-chip features 32 processing cores and contains 23 billion transistors — roughly the same number packed into our z16 chip. 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."ed to rival designs.
"It’s designed for deep learning and can be programmed to run any type of deep learning task, whether that’s processing spoken language or words and images on a screen," IBM claims of the AIU. "Our complete system-on-chip features 32 processing cores and contains 23 billion transistors — roughly the same number packed into our z16 chip. 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."
While IBM is eager to show the part off, though, it's remaining relatively quite about exact specifications, performance gains, and — even more importantly — the price, stating only that it "hope[s] to soon share news about its release."