Summary
OpenAI President Greg Brockman and Broadcom CEO Hock Tan discuss their debut custom chip, Jalapeño, designed for LLM inference. Brockman highlights significant performance-per-watt improvements and rapid nine-month design-to-tapeout timeline. Hock Tan says the chip validates Broadcom's business model of building custom AI silicon for frontier model makers. The conversation touches on industry demand for compute and the chip's unique niche.
- OpenAI and Broadcom unveil custom chip Jalapeño for LLM inference.
- Chip targets substantial improvement in performance per watt and per dollar.
- Design to tape out completed in just nine months, described as blazing fast.
- Hock Tan says this validates Broadcom's custom AI silicon business model.
- Chip is built around OpenAI's knowledge of LLM workloads and industry trends.
- Detailed technical benchmark reports promised in the coming months.