Nvidia Nemotron Passes Agentic Chip Design Tasks at 97.1% Rate, Best of Open Models
Nvidia’s Nemotron 3 Ultra model achieved a 97.1% pass rate on RTL coding tasks for chip design, outperforming other open-weight models on the tested metrics.
The system iteratively writes code to define logic, runs it through a simulator to identify failures, and rewrites the output. Nvidia tested the process across nine categories of real-world design work, averaging 6,629 tokens per iteration.
From the sources (3 posts)
@nvidiaaiWe tested Nemotron 3 Ultra on agentic chip design. The task is RTL coding: a model iteratively writes the code that defines a chip's logic, runs it through a simulator, reads the failures, and rewrites. Across nine categories of real de
@ctnzrRT @NVIDIAAI: We tested Nemotron 3 Ultra on agentic chip design. The task is RTL coding: a model iteratively writes the code that defines…
@josephjacks_RT @NVIDIAAI: We tested Nemotron 3 Ultra on agentic chip design. The task is RTL coding: a model iteratively writes the code that defines…