OpenAI Introduces Jalapeño, an AI chip designed by a LLM. 3.6x faster than Nvidia’s GB300
OpenAI announced a new AI accelerator named Jalapeño which achieves 3.6x lower inference latency than the Nvidia GB300. The new chip was developed by less than 100 engineers over the course of 20 months leveraging OpenAI’s language models to optimize circuit designs.
OpenAI recently unveiled a new AI chip called Jalapeño. Built by fewer than 100 engineers in 20 months, this chip reduces inference latency by 3.6x compared to the Nvidia GB300. They used their language models to help optimize circuits for Jalapeño.
TLDR: With the help of LLMs, they were able to improve the performance of a compute kernel from 0.31% to 88.94% of peak within ~40 hours. This is not “self-designing chips,” but rather automating away the most difficult aspects of parameter optimization.
Source: spectrum.ieee.org
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles