The way computers work today creates processing bottlenecks; they typically only process data on one hardware type at a time. But researchers from the University of California, Riverside argue there's a way to fundamentally change this, allowing multiple parts to process the same information simultaneously.This could dramatically improve computer efficiency without having to add more hardware to existing devices—and potentially change computing forever.In their research paper, UC Riverside Associate Professor Hung-Wei Tseng and Ph.D. candidate Kuan-Chieh Hsu find that current computers are "underutilizing" their potential processing power. They show how a new model, dubbed "Simultaneous and Heterogenous Multithreading" (SHMT), allows computers to engage in a new type of parallel processing where different hardware components can process the same data at the same time.While some types of parallel computing, CPU multithreading, and parallel processing already exist, SHMT departs from this because it combines the efforts of multiple different types of chips on one system. Tseng and Hsu were able to get an embedded system platform to use its CPU, Nvidia graphics card, and Tensor Processing Unit simultaneously. They found that deploying SHMT was 1.95 times faster and used 51% less energy compared to using the GPU alone. "You don’t have to add new processors because you already have them," Tseng says.Further investigation is still needed to determine how best to optimize code and hardware for such use, however."Carelessly using heterogeneous hardware components simultaneously can lead to unwanted execution results," Tseng and Hsu say in their report, pointing out that different types of hardware can generate different results in different formats from each other.If tuned for mass-market use, SHMT could have a number of impacts on the broader industry and environment. Increased device efficiency could allow organizations and individuals to save money on computers, speed up AI development while reducing its sky-high energy use, improve mobile phone and desktop computer performance, and improve cloud computing infrastructure, to name a few possibilities.
Wir nutzen Cookies auf unserer Website. Einige von ihnen sind essenziell für den Betrieb der Seite, während andere uns helfen, diese Website und die Nutzererfahrung zu verbessern (Tracking Cookies). Sie können selbst entscheiden, ob Sie die Cookies zulassen möchten. Bitte beachten Sie, dass bei einer Ablehnung womöglich nicht mehr alle Funktionalitäten der Seite zur Verfügung stehen.
The way computers work today creates processing bottlenecks; they typically only process data on one hardware type at a time. But researchers from the University of California, Riverside argue there's a way to fundamentally change this, allowing multiple parts to process the same information simultaneously.This could dramatically improve computer efficiency without having to add more hardware to existing devices—and potentially change computing forever.In their research paper, UC Riverside Associate Professor Hung-Wei Tseng and Ph.D. candidate Kuan-Chieh Hsu find that current computers are "underutilizing" their potential processing power. They show how a new model, dubbed "Simultaneous and Heterogenous Multithreading" (SHMT), allows computers to engage in a new type of parallel processing where different hardware components can process the same data at the same time.While some types of parallel computing, CPU multithreading, and parallel processing already exist, SHMT departs from this because it combines the efforts of multiple different types of chips on one system. Tseng and Hsu were able to get an embedded system platform to use its CPU, Nvidia graphics card, and Tensor Processing Unit simultaneously. They found that deploying SHMT was 1.95 times faster and used 51% less energy compared to using the GPU alone. "You don’t have to add new processors because you already have them," Tseng says.Further investigation is still needed to determine how best to optimize code and hardware for such use, however."Carelessly using heterogeneous hardware components simultaneously can lead to unwanted execution results," Tseng and Hsu say in their report, pointing out that different types of hardware can generate different results in different formats from each other.If tuned for mass-market use, SHMT could have a number of impacts on the broader industry and environment. Increased device efficiency could allow organizations and individuals to save money on computers, speed up AI development while reducing its sky-high energy use, improve mobile phone and desktop computer performance, and improve cloud computing infrastructure, to name a few possibilities.