Parallel Computing and Optimization Techniques

Modern processors stopped getting faster in isolation around two decades ago, so squeezing more performance out of hardware now means running many computations at the same time across dozens, hundreds, or even thousands of cores. Researchers working at the intersection of parallel computing and computer architecture study how to design processors, memory systems, and interconnects so that this simultaneous work proceeds efficiently without threads stalling, power budgets ballooning, or data moving more than necessary between chips and memory. GPU computing has pushed these questions to an extreme scale, exposing deep tensions between raw throughput and the energy cost of moving data, while multicore CPUs demand careful coordination to avoid bottlenecks that erase the gains parallelism promises. Open directions include building simulation platforms accurate enough to guide next-generation hardware decisions before silicon is fabricated, and finding principled ways to manage power dynamically so that performance per watt continues to improve as transistor scaling slows.

Works

205,451

Total citations

2,330,511

Keywords

Parallel ComputingPerformance OptimizationGPU ComputingMulticore ArchitecturesMemory SystemsBenchmarking