Towards Affordable, Adaptive and Automatic GNN Training on CPU-GPU Heterogeneous Platforms
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912699389050880 |
|---|---|
| author | Qiao, Tong Zhou, Ao Qi, Yingjie Wang, Yiou Wan, Han Yang, Jianlei Hu, Chunming |
| author_facet | Qiao, Tong Zhou, Ao Qi, Yingjie Wang, Yiou Wan, Han Yang, Jianlei Hu, Chunming |
| contents | Graph Neural Networks (GNNs) have been widely adopted due to their strong performance. However, GNN training often relies on expensive, high-performance computing platforms, limiting accessibility for many tasks. Profiling of representative GNN workloads indicates that substantial efficiency gains are possible on resource-constrained devices by fully exploiting available resources. This paper introduces A3GNN, a framework for affordable, adaptive, and automatic GNN training on heterogeneous CPU-GPU platforms. It improves resource usage through locality-aware sampling and fine-grained parallelism scheduling. Moreover, it leverages reinforcement learning to explore the design space and achieve pareto-optimal trade-offs among throughput, memory footprint, and accuracy. Experiments show that A3GNN can bridge the performance gap, allowing seven Nvidia 2080Ti GPUs to outperform two A100 GPUs by up to 1.8X in throughput with minimal accuracy loss. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_07421 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Towards Affordable, Adaptive and Automatic GNN Training on CPU-GPU Heterogeneous Platforms Qiao, Tong Zhou, Ao Qi, Yingjie Wang, Yiou Wan, Han Yang, Jianlei Hu, Chunming Distributed, Parallel, and Cluster Computing Graph Neural Networks (GNNs) have been widely adopted due to their strong performance. However, GNN training often relies on expensive, high-performance computing platforms, limiting accessibility for many tasks. Profiling of representative GNN workloads indicates that substantial efficiency gains are possible on resource-constrained devices by fully exploiting available resources. This paper introduces A3GNN, a framework for affordable, adaptive, and automatic GNN training on heterogeneous CPU-GPU platforms. It improves resource usage through locality-aware sampling and fine-grained parallelism scheduling. Moreover, it leverages reinforcement learning to explore the design space and achieve pareto-optimal trade-offs among throughput, memory footprint, and accuracy. Experiments show that A3GNN can bridge the performance gap, allowing seven Nvidia 2080Ti GPUs to outperform two A100 GPUs by up to 1.8X in throughput with minimal accuracy loss. |
| title | Towards Affordable, Adaptive and Automatic GNN Training on CPU-GPU Heterogeneous Platforms |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2511.07421 |