A software modification that has the potential to significantly accelerate computer processing rates was revealed by University of California Riverside (UCR) researchers in a ground-breaking study that was presented at MICRO 2023. Simultaneous and heterogeneous multithreading (SHMT) is their framework that promises to greatly increase the power and efficiency of current processors found in PCs, smartphones, and other devices.
Hardware accelerators for AI and machine learning as well as GPUs are among the many components found in modern computers. Nevertheless, these elements frequently function separately, resulting in data flow obstructions. In contrast to conventional programming approaches, SHMT enables several components to handle data at once, reducing idle resources and enhancing processing speed.
Hung-Wei Tseng, UCR associate professor of electrical and computer engineering and co-lead author of the study, emphasized that SHMT leverages existing processors, eliminating the need for additional hardware. The framework divides computational functions among different components, enabling parallel processing and maximizing efficiency.
SHMT’s innovative approach lies in its use of virtual operations (VOPs) and quality-aware work-stealing (QAWS) scheduling policy. This allows CPUs to offload functions to virtual hardware devices, ensuring workload balance and quality control.
To test the concept, the researchers built a system using NVIDIA’s Jetson Nano module, containing a quad-core ARM Cortex-A57 processor, 128 Maxwell architecture GPU cores, and a Google Edge TPU. The system demonstrated a remarkable 1.95X speed boost and a 51% reduction in energy consumption compared to baseline methods.
The implications of SHMT are significant. Not only could it enhance the performance of existing devices, but it could also reduce the need for expensive components, leading to more affordable and energy-efficient devices. For data centers, SHMT could optimize energy use and reduce carbon emissions and water consumption.
While further research is needed to refine SHMT’s implementation and determine its optimal applications, the initial results are promising. With its potential to transform computing efficiency, SHMT represents a major leap forward in the field of computer processing.