The Role of Software Optimization in Improving Computer Hardware Performance

Abstract

Optimizing software determines whether the advanced computer hardware meets its intended performance. Most processors, GPUs, accelerators, memories, and interconnects offer significant parallel capability, but if the software is not structured to make use of those resources, they may not be fully utilized. The paper focuses on a narrative review of foundational and recent scholarly research and importance of software optimisation in enhancing computer hardware. This will include discussion of compiler optimisation, algorithm improvement, parallel programming, memory and cache optimisation, profiling, and hardware–software co-design. The review identifies software optimisation techniques that enhance the performance of software by eliminating "useless computation," increasing locality, increasing parallelism, and aligning the execution pattern to hardware capabilities. But optimization is not always workload and architecture independent: A particular optimization may have a negligible effect on an application or may even harm it. Also, Amdahl's Law demonstrates that parts of programs that are executed sequentially limit the advantages of parallel hardware. The paper argues that software optimization is not just a side-benefit of hardware improvements, but a requirement to fully achieve hardware performance in today's computing systems.