Investigating the Memory-Bandwidth Bottleneck in HPC Applications Using Sparse Data Structures

As HPC systems and applications continue to get more optimized, a bottleneck is forming around memory-bandwidth. A significant factor in this bottleneck is the large memorybandwidth burden of sparse matrix operations, which are a common type of operation in modern HPC workloads. Therefore, improvements in the memory-bandwidth burden of sparse matrix operations would lead to significant improvements in the speed of HPC applications. I intend to pursue such improvements in my PhD.