Poster Type: Research Posters
Author: Grace McLewee (University of Oregon), Mohammad Alaul Haque Monil (Oak Ridge National Laboratory (ORNL)), Pedro Valero-Lara (Oak Ridge National Laboratory (ORNL)), Narasinga Miniskar (Oak Ridge National Laboratory (ORNL)), William Godoy (Oak Ridge National Laboratory (ORNL)), Allen Malony (University of Oregon), Keita Teranishi (Oak Ridge National Laboratory (ORNL)), Jeffrey Vetter (Oak Ridge National Laboratory (ORNL))
Abstract: The growing heterogeneity of HPC programming models and hardware architectures has made automated performance modeling more essential and more difficult. Compiler-based frameworks such as COMPASS and MAPredict need programming model-specific integration; frontier LLMs are flexible but predict runtime poorly; custom LLMs buy accuracy at prohibitive cost. We present GenASPEN, an agentic AI framework that assigns each task to the component best equipped for it: a Code Translation Agent (a context-aware chain-of-thought pipeline over GPT-OSS 120B, or a Code Llama 7B assistant fine-tuned with ChatHPC) converts CUDA/HIP kernels into ASPEN application models, and a Performance Modeling Agent combines them with machine and analytical models to predict FLOPs, loads, stores, and execution time on modern GPUs. Across 22 benchmarks on NVIDIA H100 and AMD MI300A, GenASPEN reaches 98.5% operation-count and 94.40%/93.09% execution-time accuracy, beating four frontier LLMs on 16 of 22 kernels with zero manual annotation.
Best Poster Finalist (BP): yes
Poster: PDF
Poster Summary: PDF