Cerebras Touts CS-4 Chip with Up to 30x Faster AI Inference than Nvidia GPUs
Cerebras has unveiled its latest wafer-scale engine, CS-4, which promises to deliver up to 30 times faster AI inference compared to NVIDIA's next-generation GPU racks. The company made this bold claim at its Supernova 2026 event in San Francisco on August 18.
The CS-4 chip is designed to keep compute and memory close together, reducing the need for high-bandwidth memory and minimizing memory movement. This approach allows Cerebras to deliver faster performance without sacrificing power efficiency.
Investors Business Daily reported that the CS-4 system delivers 750 petaflops of AI compute, 7.2 terabits per second of I/O bandwidth, and 129.6 petabytes per second of memory bandwidth. The company claims it can deliver up to twice the speed of its previous model, CS-3, and up to 10 times more throughput per watt.
Cerebras is positioning itself as a competitor to NVIDIA's dominance in AI hardware. While NVIDIA remains the default choice for training, Cerebras aims to prove that its wafer-scale systems are the better option for inference tasks. The company has already secured several high-profile customers, including Mistral and G42.
NVIDIA is not sitting idle, however. Its next-generation GPU racks will offer improved performance and power efficiency, making it a tough competitor for Cerebras to beat.