AI start-up unveils Cerebras Wafer-Scale Engine (WSE), the largest chip ever built
Cerebras Systems says its, “56x larger than any other chip, the WSE delivers more compute, more memory, and more communication bandwidth. This enables AI research at previously-impossible speeds and scale”.
Transistor record
The firm has packed in 1.2 trillion transistors on the chip, leagues ahead of the 32 billion transistors that AMD built into a recent process.
There is also a fundamental difference in the construction of the chip. Previous chips comprised of more chips built into the top of a 12-inch silicon wafer. The WSE device is a single interconnected-chip on a single wafer.
Made for artificial intelligence
Though a Samsung flash memory chip had 2 trillion transistors, the WSE is a processor with 400,000 cores across 42,225 mm². Further, it boasts of 3,000 times more high-speed, on-chip memory and 10,000 times more memory bandwidth.
The speed and processing muscle of the WSE make it ideal for AI applications. After all, applications require chips to produce information quicker and generate more answers via lower “training time.”
“Every architectural decision [will] optimize performance for AI work,” said Andrew Feldman, co-founder, and CEO of Cerebras Systems. “The result is that the Cerebras WSE delivers, depending on workload, hundreds or thousands of times the performance of existing solutions at a tiny fraction of the power draw and space.”