AI Generated Image
OpenAI unveils its first custom chip, built by Broadcom
TechCrunch • • business
Key Points:
- OpenAI unveiled its first custom-built inference processor, Jalapeño, developed in collaboration with Broadcom and designed specifically for OpenAI’s AI inference systems, with assistance from OpenAI’s own AI models.
- Early testing shows Jalapeño offers significantly better performance-per-watt compared to current alternatives, aiming to reduce OpenAI’s reliance on Nvidia GPUs for inference workloads.
- The chip focuses on inference tasks, such as running pre-built AI models in real time, particularly benefiting applications like coding models, while more intensive tasks like pre-training will likely still use Nvidia hardware.
- OpenAI’s approach integrates chip design with its broader infrastructure, optimizing across hardware and software layers to improve speed, reliability, and cost-efficiency of its AI models and products.
- This move aligns OpenAI with other tech giants like Google and Amazon, who have also developed custom AI accelerators to enhance machine learning workload performance and economics.
