ARTICLE AD BOX
New Delhi: In a significant new development for OpenAI, the company is launching its first custom AI inference chip in collaboration with Broadcom called “Jalapeño.” The new processor is optimized to run large language models (LLMs) more efficiently, potentially with ChatGPT and other services, while minimizing external chip dependence.
The name Jalapeño suggests that it is an “Intelligence Processor” that adheres to OpenAI’s vision of the future of AI inference. Inference refers to the stages of a process after an AI model is trained, when it is used to generate responses to requests from a user. The ability to make inferences quickly and effectively has become a crucial area of focus for AI firms, given the millions of people who are witnessing the technology every day. OpenAI and Broadcom say the chip went from design to production phase in just nine months. The companies attributed some of the acceleration of the chip development process to OpenAI’s own AI models, which accelerated certain aspects of the work and enabled engineers to work more quickly than on other semiconductor projects.
The chip is designed specifically for large language models, and initial testing has demonstrated good performance. Users might expect to see lower operating costs and reduced data centre power consumption through it, too, because, according to OpenAI, Jalapeño performs better than numerous existing AI processors per Watt. “The idea behind Jalapeño is to oppose some of the top AI accelerators in the market, such as Nvidia and Google products,” said Broadcom CEO Hock Tan. TSMC is making the processor, and Canadian tech company Celestica is doing the system integration.
The launch is also a sign that OpenAI is looking to take a step to have more control over the ground-level infrastructure associated with its AI products. Like several tech companies, such as Google, Amazon, and Meta, OpenAI also has a portfolio of custom silicon upgrades to enhance performance and move away from third-party hardware. OpenAI insists Jalapeño is just the first of many fun times ahead. The company is developing an AI hardware architecture that will eventually be even more capable and efficient than the chips that are being developed by Broadcom. The deployment of Jalapeño throughout OpenAI’s infrastructure is anticipated to start later this year and aid in surmounting the increasing demand for AI services worldwide.
The release of Jalapeño is a significant development in OpenAI’s journey towards enhancing its artificial intelligence capabilities. The company’s home-grown AI processor promises to enhance user experience with faster, more reliable, and more universally available AI experiences, as well as boost its market entry in the rapidly changing AI hardware segment.
English (US) ·