OpenAI and Broadcom unveil LLM-optimized inference chip
4 hours ago
- #AI Hardware
- #Full-Stack AI
- #LLM Inference
- OpenAI and Broadcom have introduced Jalapeño, a first-generation LLM-optimized inference chip.
- Early testing indicates Jalapeño offers superior performance per watt compared to current top accelerators.
- The chip was developed in nine months, aided by OpenAI's own AI models to accelerate design.
- Jalapeño is designed from scratch for LLM inference, tailored to current and future industry models.
- It supports OpenAI's full-stack strategy, optimizing from chip to product for efficiency and accessibility.
- The platform aims for gigawatt-scale deployment with partners like Microsoft starting in 2026.
- Jalapeño enhances compute efficiency, driving better AI models, products, and broader democratization of AI.