Cerebras Launches Fourth-Generation AI Inference System CS-4, Doubling Performance and Power Consumption

Cerebras has launched its fourth-generation AI inference system, the CS-4, which is based on a 5nm WSE-3 wafer. It achieves double the performance by doubling the clock frequency and power consumption. A single CS-4 cabinet houses three wafers and features a modular design with a TDP of approximately 125 to 135 kW. The CS-4 offers an inference speed of nearly 4000 tokens per second per user, about twice that of the CS-3, and supports decomposed inference with heterogeneous systems such as AMD and AWS Trainium. Cerebras claims that the CS-4 provides approximately 2000 times the on-chip memory bandwidth of NVIDIA's Rubin (43 PB/s), although the 44 GB SRAM capacity remains unchanged, necessitating multi-wafer stacking for long-context inference. For example, with DeepSeek V4 Pro (1.6T parameters), around 20 systems are needed for a 1M context window, and about 40 systems for 256 concurrent users, corresponding to a CAPEX exceeding $20 million. Cerebras is collaborating with clients like OpenAI, aiming for an annual performance improvement of about 2 times, with a target of achieving a 20-fold throughput increase by 2027. The cabinet design of the CS-4 will carry over to the next-generation Nexus platform.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.