Startup microchip developer Cerebras on Monday unveiled the largest computer processing chip ever built. The chip called “Wafer-Scale Engine” (WSE) was designed for the company’s deep learning machine built to power AI systems. Most chips that are developed today can fit on the tip of your finger, Cerebras chip, on the other hand, measures 8 1/2 inches on each side.
The WSE’s surface contains 400,000 cores with 1.2 trillion transistors and compared to the Nvidia V100 chip, the WSE has 3,000 times the memory circuits.
The new WSE offers a couple of key advantages. The first and obvious one is computational power, with so many processing cores and chip memory the chip can handle thousands of models in the time it takes a conventional systems to handle just one. The other problem the WSE chip solves is there is no lag time between cores. Current servers with multiple chips slow down when sending data between the chips over the slower wires linking them on a circuit board.
Cerebras new chip will help to solve complex AI problems faster, cheaper, and use far less energy than current systems.