A large AI chip with 2.6 trillion transistors was released.

Cerebras Systems, a company that manufactures chips for machine learning and deep learning, announced that it has developed Wafer Scale Engine 2 (WSE2), a computer chip with 2.6 trillion transistors mounted on a wafer.

WSE2 is equipped with 2.6 trillion transistors, 950,000 cores, and 40GB of on-chip memory in an area of 20×22cm. In 2019, the company developed WSE, a chip with 1.2 trillion transistors, but WSE2 has more than doubled its performance.

The biggest change in WSE is the manufacturing process. WSE2 uses TSMC 7nm manufacturing process to double the transistor density, increasing the speed by 40% and reducing power consumption by 60%. The reason why the size of the chip itself has not changed in WSE is that it is the limit size that can be mass-produced, he explains. Celebras Systems said that a large chip is important for the processing speed of AI computing information, and it is planning to provide the new WSE to various customers. Related information can be found here.