Search papers, labs, and topics across Lattice.
This paper introduces D2K-Bench, a novel benchmark comprising 26 tasks and 85 workloads, to systematically evaluate LLM agents' ability to translate expert design guidance into efficient GPU kernels. The benchmark assesses guidance across algorithmic, dataflow, and low-level optimization levels, revealing that expert guidance significantly improves correctness from 93.1% to 98.5% and boosts performance scores from 1.46 to 1.95 across 26 tasks. Key findings highlight the substantial impact of design guidance on performance, while also pinpointing specific design properties that LLM agents currently struggle to implement.
Expert design guidance nearly doubles LLM agent speedups for GPU kernel generation, yet agents still miss critical low-level optimizations.
GPU kernels generated by large language model (LLM) agents can remain less efficient than expert implementations, but runtime alone does not reveal how the gap relates to design discovery and implementation. We introduce D2K-Bench, a diagnostic benchmark of 26 tasks and 85 workloads that measures how effectively agents translate expert design guidance into efficient GPU kernels. The guidance covers L1: high-level algorithmic insights, L2: dataflow design, and L3: low-level optimization tricks, including dependencies among these levels. Pairwise runs with and without guidance share task descriptions, workloads, tools, hardware, and a 350-turn budget. Complementary assessments examine independently proposed designs and the design properties implemented in generated code. Across five models on NVIDIA B200 GPUs, guidance raises correctness over 130 model-task pairs from 93.1% to 98.5% and increases the Performance Score over all 26 tasks from 1.46 to 1.95. For the three frontier models with correct submissions on all 26 tasks in both runs (GPT-6-Astra, Claude-Opus-4.8, and GPT-5.6-Sol), geometric mean speedup increases from $1.69\times$ to $2.49\times$. Across all five models, the mean combined implementation score increases from 57 to 70 out of 100. These results show the value of expert design guidance while identifying design properties that remain unimplemented.
