Cerebras Systems Inc. Class A Common StockCerebras unveils CS-4 with new architecture, doubling performance and boosting inference speed, driving growth.
Cerebras Systems unveiled a new rack-scale AI system called CS-4, which it says can run large models dramatically faster than traditional GPU infrastructure. The CS-4 combines three WSE-3 Turbo processors and is the first system built around the company's next-generation Nexus rack-scale architecture, delivering up to twice the performance of its CS-3 predecessor. On the GPT-OSS-120B model, CS-4 reportedly produced more than 4,400 tokens per second per user, which Cerebras says is as much as 30 times faster than GPU-based alternatives, with up to 10 times higher throughput per watt than CS-3. First shipments are expected this quarter. The launch comes as Cerebras' inference business accelerates, with second-quarter cloud and services revenue jumping 281% year over year to $126 million, while core total revenue reached $209.9 million, up 103%, and the company ended June with $25.4 billion in remaining performance obligations.
Cerebras Systems Inc. Class A Common StockCerebras unveils CS-4 with new architecture, doubling performance and boosting inference speed, driving growth.
NVIDIA CorporationCerebras claims CS-4 is up to 30x faster than GPU-based alternatives, directly challenging NVIDIA's dominance in AI inference.
Advanced Micro Devices IncCerebras claims CS-4 is up to 30x faster than GPU-based alternatives, directly challenging AMD's data center offerings.