CB-SpMV: A Data Aggregating and Balance Algorithm for Cache-Friendly Block-Based SpMV on GPUs

Researchers introduced CB-SpMV, a GPU SpMV method that improves cache locality and load balancing.

It uses adaptive 2D blocking, data aggregation, block-aware column grouping, and inter-block load balancing.

Across 2,843 SuiteSparse matrices, it achieved higher cache hit rates and up to 3.95x speedup over leading baselines.

The results suggest a practical new route to faster sparse linear algebra on modern GPUs.