Back to all articles
•Technology

OpenAI and Broadcom Unveil 'Jalapeño': A Custom AI Inference Chip to Enhance Performance and Efficiency

View original source

OpenAI, in collaboration with Broadcom, has announced the release of its first custom-built inference processor, named Jalapeño. This development was officially revealed on Wednesday and marks a significant step towards reducing OpenAI's reliance on third-party hardware, notably Nvidia's GPUs. Key points include:

  • Collaboration Details: While the partnership with Broadcom was officially announced in October, OpenAI's plans for custom chip design have been a topic of industry speculation for some time.
  • Performance Insights: The early testing of Jalapeño indicates a noteworthy improvement in performance-per-watt compared to current state-of-the-art alternatives.
  • Strategic Goals: The chip aims to optimize the inference process within AI models by lowering operational costs during real-time coding tasks, potentially boosting OpenAI's economic scalability.
  • Strategic Vision: As stated by OpenAI's president, Greg Brockman, the development of Jalapeño is an effort to target underserved workloads and accelerate potential advancements in AI capabilities.
  • Long-term Impact: The chip is a part of OpenAI's strategy to manage its infrastructure—from chip architecture and memory systems to networking and product experience—all focused on making AI models more efficient, reliable, and cost-effective for users.

Through customized chip development, OpenAI aligns its entire technology stack towards enhancing the operational and economic efficiency of its AI models.