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

Figure 9

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