OpenAI and Broadcom reveal Jalapeño chip to optimize large language model inference (opens in new tab)

OpenAI and Broadcom have unveiled Jalapeño, a custom-designed intelligence processor specifically architected for large language model (LLM) inference. This first-generation accelerator was developed from initial design to production in just nine months, a cycle significantly accelerated by using OpenAI’s own models. The chip is intended to provide a full-stack infrastructure solution that balances compute, memory, and networking resources to achieve high utilization. <a href="