A Novel CSR-Based Sparse Matrix-Vector Multiplication on GPUs

Figure 10

Parallel efficiency of PCSRI and PCSRII with communication on four GPUs.