Cerebras unveils CS-4 with double the performance on the same chip
decoder ·
Cerebras has announced its CS-4 AI accelerator, which CEO Andrew Feldman claims is the fastest system in the industry. The CS-4 is a rack-scale product, arriving as a complete server cabinet with compute, power, and cooling. It uses the same 5nm WSE-3 chip as the CS-3 but doubles performance through increased clock speed, power, and cooling. Each rack now contains three wafers instead of two and can deliver up to 4,400 tokens per second per user, which Cerebras says is up to 30 times faster than Nvidia GPU-based systems. Memory capacity remains unchanged at 44 GB per wafer. The new modular 'Backpack' design enables faster assembly and supports disaggregated inference through partners like AMD and AWS Trainium. Analysts at SemiAnalysis noted that the networking improvements are relatively small, and more details will be shared at the Hot Chips conference. Cerebras hardware is already used by OpenAI for Codex Spark, among other applications.
decoder