Cerebras introduces CS-4 AI accelerator to provide 30x faster inference speeds than traditional GPU systems
Cerebras has unveiled the CS-4, a new rack-scale AI accelerator designed to deliver up to 30 times faster inference speeds than production GPU systems. The system utilizes three Wafer Scale Engine 3 Turbo processors, which are the largest AI semiconductors ever built, featuring 4 trillion transistors. By leveraging static random-access memory (SRAM) on dinner-plate-sized wafers, the CS-4 achieves high-speed performance on large frontier models without sacrificing model capability. Cerebras CEO Andrew Feldman stated that the CS-4 fundamentally changes the AI paradigm by delivering industry-leading speeds on the largest models. The new architecture, known as the Nexus Platform Architecture, optimizes compute, power, and I/O to reduce latency and increase throughput. While Cerebras shares have seen a recent decline in stock price and a reported loss per share in Q2, the company continues to offer AI cloud services and remains a competitive force against industry leaders like Nvidia. The first shipments of the CS-4 begin this quarter.