Cerebras says its AMD GPU tag team is live, delivering 5× throughput per megawatt versus CS-3 alone
Cerebras says its live inference setup pairs AMD GPUs for prompts with wafer-scale systems for responses.
TLDR
AMD handles the prompts; Cerebras handles the replies, according to Cerebras. Its production deployment pairs AMD Instinct GPUs with CS-3 systems, and the company reports five times the throughput per megawatt of CS-3 alone without reducing token-generation speed. Cerebras plans broader deployment across hundreds of megawatts in its cloud and partner infrastructure by the end of 2027.
Combined views
—
2 Sources, first seen ago

Comments