Skip to main content

This is the world's largest computer chip

(Image credit: Shutterstock.com)

A new AI startup has launched what it says is the largest semiconductor chip ever built.

Cerebras Systems has taken the wraps off its Wafer Scale Engine (WSE), which packs in 1.2 trillion transistors, as a new type of computer optimised exclusively for AI and deep learning capabilities.

This puts it far ahead of other chips currently on the market, with recent launches from other chip leaders such as AMD and Nvidia packing in between 20 and 30 billion transistors.

Enormous

Cerebras says that its hardware will benefit AI processing capabilities, which are often held back by a lack of available compute capacity.

The company says testing a single new hypothtesis or training a new model is both expensive and time-heavy, often taking weeks or even months on currently-available systems. 

Cerebras believes its new chip can help meet the "enormous computational demands" of deep learning, being purpose-built to allow for greater computation than any other hardware on the market today. 

(Image credit: Cerebras)

Unlike many other high-powered chips on the market today, which are actually a collection of chips piled on top of a silicon wafer, the WSE is a single chip on a single wafer, allowing for much faster transfer speeds and processing capacity.

The WSE is more than 56X larger than the largest graphics processing unit, containing 3,000X more on-chip memory and capable of achieving more than 10,000X the memory bandwidth, with high-speed memory close to each core ensuring that cores are always occupied doing calculations. 

Across its 46,225 square millimeters of silicon, (making it slightly smaller than a typical tablet) the chip is able to deliver 400,000 AI-optimised cores, and its specialised memory architecture ensures each of these cores operates at maximum efficiency. 

The WSE provides 18 gigabytes of fast, on-chip memory distributed among the cores in a single-level memory hierarchy, one clock cycle away from each core. 

There's no news yet on when the hardware will be available for wider use, with Cerebras simply saying updates will be coming soon.