IBM Research: Open Models Cut AI Chip Design Inference Cost by 11.5x
IBM demonstrates that open frontier models with llm-d reduced inference costs for quantum error correction hardware optimization by 11.5x compared to proprietary configurations.
ImportanceMaterialEvidenceE2 unreplicatedWrite-upQuick
IBM Research demonstrated that using open frontier models combined with llm-d distributed inference can reduce the cost of AI-driven chip design exploration by 11.5 times compared to proprietary configurations.
In a test optimizing quantum error correction decoders, the open-model setup achieved a 52x speedup for approximately $26 in inference costs. A proprietary model setup cost $299 for the same search budget, though it achieved a higher 204x speedup.
These results are from a single search run per configuration and characterize observed performance rather than establishing a universal model ranking.