English

CUCo: An Agentic Framework for Compute and Communication Co-design

Distributed, Parallel, and Cluster Computing 2026-03-04 v1 Hardware Architecture Machine Learning Multiagent Systems

Abstract

Custom CUDA kernel development is essential for maximizing GPU utilization in large-scale distributed LLM training and inference, yet manually writing kernels that jointly leverage both computation and communication remains a labor-intensive and error-prone process. Prior work on kernel optimization has focused almost exclusively on computation, leaving communication kernels largely untouched even though they constitute a significant share of total execution time. We introduce CUCo, a training-free agent-driven workflow that automatically generates high-performance CUDA kernels that jointly orchestrate computation and communication. By co-optimizing these traditionally disjoint components, CUCo unlocks new optimization opportunities unavailable to existing approaches, outperforming state-of-the-art baselines and reducing end-to-end latency by up to 1.57×1.57\times.

Keywords

Cite

@article{arxiv.2603.02376,
  title  = {CUCo: An Agentic Framework for Compute and Communication Co-design},
  author = {Bodun Hu and Yoga Sri Varshan and Saurabh Agarwal and Aditya Akella},
  journal= {arXiv preprint arXiv:2603.02376},
  year   = {2026}
}
R2 v1 2026-07-01T11:00:01.042Z