Researcher discovers innovation to double computer speed for free – Computer

A researcher claims to have discovered an innovative method that can double the speed of computers. Without any additional cost in hardware.

This method, called a heterogeneous simultaneous multi-threading (SHMT) architecture, is described in a paper co-authored by Hong Wei Cheng, an assistant professor of electrical and computer engineering at the University of California, Riverside, and Kuan-Chih Hsu, a graduate student at the University of California, Riverside. , Riverside. computer science.

The SHMT framework currently runs on a platform that simultaneously uses an ARM multi-core processor, an Nvidia graphics card, and a Tensor hardware accelerator. In tests, the system recorded a 1.96 times speed increase and a 51% reduction in power consumption.

Modern computers increasingly include graphics processing units and hardware accelerators for artificial intelligence and machine learning, or DSPs, as their core components, Cheng explained. But these components process information separately, creating “bottlenecks,” that is, delays in the flow of information. The goal with SHMT is to address this problem by allowing individual components to work simultaneously, thus enhancing processing efficiency.

The implications of this discovery are significant. Not only will this reduce the cost of equipment, but it will also reduce carbon dioxide emissions from generating the energy needed to run servers in large data centres. Moreover, it can also reduce water needs for cooling servers.

Speaking to techradar, Cheng said that the SHMT framework, if adopted by Microsoft in a future version of Windows, could provide a free performance boost to users. The research's energy saving claim is based on the rationale that by reducing the process execution time, less energy is consumed, even when using exactly the same devices.

Of course, there is a footnote (as always). In his conclusions, Cheng notes that more research is needed to answer questions about system implementation, hardware support, code optimization, and which applications can reap the greatest benefits.

Although no engineering effort is necessarily needed, Cheng says “we would definitely need changes in the operating system (such as OS drivers) as well as programming languages ​​(such as Tensorflow/PyTorch)” to make the proposed method work.

The study, presented at the 56th Annual IEEE/ACM International Symposium on Microarchitecture in Toronto, Canada, was recognized by the Institute of Electrical and Electronics Engineers (IEEE) as one of 12 studies to be included in the Institute's “Best of Studies” publication. . Selections from Computer Architecture Conferences scheduled for later this year.