OpenAI Says Its Models Cut AMD GPU Design Work From a Year to Five Months
OpenAI used its o3-mini model to help design Jalapeño, the inference chip it is co-developing with Broadcom, shrinking vector unit cell area by about 13% and matrix unit cell area by 10%.
1 source 2h ago
Reporting from 1 source: GIGAZINE.
Why it matters
OpenAI is selling its models as chip design tools, and the named customers are the companies whose silicon it also buys and builds alongside.
Key Facts
- OpenAI co-develops the inference chip Jalapeño with Broadcom.
- o3-mini reduced Jalapeño's vector unit cell area by about 13% and matrix unit cell area by 10%.
- OpenAI's work with AMD on GPU circuit code achieved up to 14% PPA improvement over a highly optimized baseline.
- AMD CEO Lisa Su commented on the chip design work.
- OpenAI and Synopsys developed GPT-Synopsys, a model for operating Synopsys EDA tools.
Work with AMD on GPU circuit code achieved up to 14% PPA improvement against an optimized baseline, and a gain OpenAI says would have taken about a year arrived in five months.
AMD is folding the changes into product designs, with PPA up by a single-digit percentage across the main blocks of its next-generation flagship CPU. AMD CEO Lisa Su said engineers at both companies are pushing the frontier of chip design.
OpenAI also tested its models against PicoRV32, a RISC-V core built for FPGAs. GPT-6.2 Astra gave the largest performance gain at the lowest cost among the five models tried. With Synopsys, OpenAI developed GPT-Synopsys, a model for operating the company's EDA tools.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google