NeutronSparse: Coordinating Heterogeneous Engines for Sparse Matrix Multiplication on NPUs

Summary: NeutronSparse identifies poor coordination among heterogeneous NPU compute units—not raw compute—as the key SpMM bottleneck. Its sparsity-aware workload balancing and locality-aware tile orchestration deliver 1.26–7.78× speedups over NPU baselines and up to 3.07× over A100 GPU libraries. (summarized by gpt-6-luna on Oct 22 2026)

Paper ID: h4b18e2e9451c7f23
Venue: SIGMOD
Year: 2027
Pagerank: 4.9605343e-05
Overall Rank: 10,404 | 30.27%
DOI: 10.1145/3837119
PDF: Download (CC BY 4.0)

Incoming Non-self Citations Over Time

No non-self incoming citations found for this paper in this database.

Authors

1. Xin Ai (Northeastern University)
2. Zeyu Ling (Northeastern University)
3. Hao Yuan (Northeastern University)
4. Qiange Wang (Northeastern University)
5. Yanfeng Zhang (Northeastern University)
6. Yutao Peng (Northeastern University)
7. Ge Yu (Northeastern University)

BibTeX Citation

@inproceedings{ai_sigmod27,
        title = {{NeutronSparse: Coordinating Heterogeneous Engines for Sparse Matrix Multiplication on NPUs}},
        author = {Ai, Xin and Ling, Zeyu and Yuan, Hao and Wang, Qiange and Zhang, Yanfeng and Peng, Yutao and Yu, Ge},
        series = {{SIGMOD} '27},
        booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
        publisher = {Association for Computing Machinery},
        doi = {10.1145/3837119},
        url = {https://dl.acm.org/doi/10.1145/3837119},
        year = {2027}
}

Incoming Citations (Sorted by Pagerank)

Showing 0 of 0 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank

Outgoing Citations (Sorted by Pagerank)

Showing 8 of 8 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
203	AliGraph: A Comprehensive Graph Neural Network Platform	2019	VLDB	0.00025234479
1,047	SANCUS: Staleness-Aware Communication-Avoiding Full-Graph Decentralized Training in Large-Scale Graph Neural Networks	2022	VLDB	0.00012313386
2,273	NeutronStar: Distributed GNN Training with Hybrid Dependency Management	2022	SIGMOD	8.7079974e-05
2,325	Scalable and Efficient Full-Graph GNN Training for Large Graphs	2023	SIGMOD	8.625677e-05
4,884	NeutronOrch: Rethinking Sample-based GNN Training under CPU-GPU Heterogeneous Environments	2024	VLDB	6.3564568e-05
6,947	HongTu: Scalable Full-Graph GNN Training on Multiple GPUs	2023	SIGMOD	5.6213095e-05
7,036	Can Graph Reordering Speed Up Graph Neural Network Training? An Experimental Study	2025	VLDB	5.5993909e-05
8,174	NeutronTP: Load-Balanced Distributed Full-Graph GNN Training with Tensor Parallelism	2025	VLDB	5.3692867e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	10,889	ThunderGNN: Unlocking Tensor Cores for Graph Neural Networks	2026	VLDB
2	7,313	Parallel Algorithms for Sparse Matrix Multiplication and Join-Aggregate Queries	2020	PODS
3	13,716	STile: Searching Hybrid Sparse Formats for Sparse Deep Learning Operators Automatically	2024	SIGMOD
4	11,113	NeutronCloud: Resource-Aware Distributed GNN Training in Fluctuating Cloud Environments	2026	VLDB
5	4,118	Fast Sparse Matrix-Vector Multiplication on GPUs: Implications for Graph Mining	2011	VLDB
6	6,733	DistME: A Fast and Elastic Distributed Matrix Computation Engine using GPUs	2019	SIGMOD
7	4,884	NeutronOrch: Rethinking Sample-based GNN Training under CPU-GPU Heterogeneous Environments	2024	VLDB
8	9,835	NeutronHeter: Optimizing Distributed Graph Neural Network Training for Heterogeneous Clusters	2026	SIGMOD
9	8,174	NeutronTP: Load-Balanced Distributed Full-Graph GNN Training with Tensor Parallelism	2025	VLDB
10	7,544	NeutronTask: Scalable and Efficient Multi-GPU GNN Training with Task Parallelism	2025	VLDB