Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
June 17, 2026IEICE Transactions on Information and SystemsOpen Access

IMAX-SpMM: An Energy-Efficient and Dataflow-Optimized SpMM Kernel on CGLA for GNNs

View Full Paper
Ask AI
Bookmark
Share

Authors

KAK AsahinaNara Institute of Science and TechnologyDKD. Y. KimNara Institute of Science and TechnologyYNYasuhiko NAKASHIMANara Institute of Science and Technology

Discussion

Loading...

Member takes

Implication

Randomized trial demonstrates improved energy efficiency and performance in Graph Neural Networks, suggesting effective data management strategies.

Key Points

  • The aim is to enhance energy efficiency and performance in GNNs through dataflow optimizations in SpMM.
  • Developed IMAX-SpMM, a dataflow-centric SpMM kernel utilizing coarse-grained linear arrays.
  • Minimized data padding and optimized memory access patterns for energy savings.
  • Experimental evaluation performed on GCN with comparisons to i9-10940X and RTX3090.
  • Achieved up to 3.07× speedup over an i9-10940X and 1.64× over an RTX3090.
  • Improved energy efficiency by up to 246.60× compared to an i9-10940X and 25.60× compared to an RTX3090.
  • Attained a data reuse rate of 60.1% for DMA-loaded data, enhancing memory transfer efficiency.

Cite This Study

Asahina et al. (2026) studied this question.

synapsesocial.com/papers/6a32398fd50b63ecad20508ahttps://doi.org/10.1587/transinf.2026pap0002
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1DistSpMM: Accelerating Sparse Matrix Dense Matrix Multiplication on GPUs2026
  2. 2NM-SpMM: Accelerating Matrix Multiplication Using N:M Sparsity with GPGPU2025
  3. 3Rethinking Tiling and Dataflow for SpMM Acceleration: A Graph Transformation Framework2025 · 4 citations
  4. 4DTC-SpMM: Bridging the Gap in Accelerating General Sparse Matrix Multiplication with Tensor Cores2024 · 46 citations
  5. 5LO-SpMM: Low-cost Search for High-performance SpMM Kernels on GPUs2024 · 7 citations