Archive — US — techcrunch
OpenAI unveils its first custom chip, built by Broadcom
OpenAI has launched its first custom-built inference processor, Jalapeño, developed in collaboration with Broadcom. This chip aims to improve performance-per-watt for AI models and reduce reliance on Nvidia's GPUs. Although still in testing, early results show promising efficiency, with the chip designed specifically for inference tasks to enhance the company's AI offerings.
Key points
- OpenAI unveiled its custom inference processor, Jalapeño, in partnership with Broadcom.
- Jalapeño is designed for OpenAI’s AI models, optimizing performance-per-watt versus existing models.
- The partnership aims to decrease dependence on Nvidia GPUs, allowing for cost reductions in AI operations.
- The chip focuses on inference tasks, crucial for real-time AI interactions, while more intensive tasks may still use Nvidia.
- OpenAI's comprehensive approach involves development across all tech layers, enhancing model performance and reliability.
Why it matters
- Custom chips are a growing trend among major tech companies to enhance machine learning performance.
- Google and Amazon have previously developed similar 'AI accelerators' aimed at increasing efficiency in AI workloads.
- The shift toward proprietary hardware reflects an ongoing move within the tech industry to optimize AI processes and reduce costs.
OpenAI's introduction of Jalapeño marks a significant step in optimizing AI infrastructure, potentially transforming the economic landscape of AI while mitigating reliance on traditional hardware providers like Nvidia.