OpenAI and Broadcom's Jalapeño, a Custom Inference ASIC: Inference ASIC vs GPU
chips gpu inference openai
| Source: Dev.to | Original article
OpenAI and Broadcom unveil a custom AI inference chip. It offers 50% cost savings compared to typical AI GPUs.
OpenAI and Broadcom have unveiled Jalapeño, a custom inference ASIC designed for AI inference. This is OpenAI's first custom-built AI chip, developed in partnership with Broadcom. The chip is built on TSMC's 3nm process and targets 50% lower cost per token than Nvidia GPUs.
This development matters as it marks a significant shift towards specialized hardware for AI inference, potentially reducing costs and increasing efficiency. By using an application-specific integrated circuit (ASIC) instead of general-purpose GPUs, OpenAI and Broadcom aim to achieve better performance and energy efficiency for large language models (LLMs) inference.
As the first custom AI chip from OpenAI, Jalapeño's performance and cost savings will be closely watched. With initial deployment expected by the end of 2026, the industry will be looking to see how this new chip stacks up against existing GPU solutions from companies like Nvidia. The success of Jalapeño could pave the way for further adoption of custom AI chips in the industry.
Sources
Back to AIPULSEN