Cerebras Systems, CS System
Cerebras CS-1 — the world’s fastest AI accelerator
The CS-1 is a system solution that consists of innovations across three dimensions: a) the Cerebras Wafer Scale Engine (WSE) — the industry’s largest and only trillion-transistor processor, b) the Cerebras System and c) the Cerebras software platform.The CS-1 is 26-inches (15 rack units) tall and fits in one-third of a standard datacenter rack. It houses the Cerebras Wafer Scale Engine and feeds the 400,000 AI-optimized compute cores and 18 Gigabytes of high-speed, on-wafer memory with 1.2 terabits per second of data. This combination of enormous input/output bandwidth — 12 x 100 Gigabit Ethernet lanes — and 18 Gigabytes of fast, on-chip memory enables the CS-1 to deliver more calculations per unit time than any machine ever built. And since all the computation and communication remains on-chip, where extraordinary power efficiency is to be had, the CS-1 consumes only a small fraction of the power and space of alternative solutions. Powering and cooling such a large chip requires fundamental technology innovations. The custom power delivery and cooling technology inside the CS-1 keeps the chip running at full performance at a temperature well below the operating temperature of traditional processors. In microelectronics, high temperatures are the enemy of reliability; the lower operating temperature of the WSE enhances both its reliability and performance. Finally, unlike clusters of graphics processing units — which can take weeks or months to set up, require extensive hyperparameter tuning to converge on, occupy dozens of datacenter racks and require proprietary InfiniBand to cluster — the CS-1 takes minutes to set up. Simply plug in the standardsbased 100 Gigabit Ethernet links to a switch, and you are ready to train models at wafer-scale speed.
The CS-1 has been deployed in some of the largest compute environments in the world, including the US Department of Energy’s supercomputing sites. CS-1s are currently being used to address some of the most difficult challenges of our time — from accelerating AI in cancer research, to better understanding and treating traumatic brain injury, to furthering discovery in fundamental science around the characteristics of black holes. With this breakthrough in performance, the Cerebras CS-1 eliminates the primary impediment to the advancement of artificial intelligence, reducing the time it takes to train models from months to minutes and from weeks to seconds, allowing researchers to be vastly more productive. In so doing, the CS-1 reduces the cost of curiosity, accelerating the arrival of the new ideas and techniques that will usher forth tomorrow’s AI.
