Scientists find a way to double PC power without adding new hardware
Researchers at the University of California, Riverside have developed a new approach to parallel data processing that has the potential to double the speed of computing and halve the power consumption of computers. This method, called simultaneous and heterogeneous multithreading (SHMT), allows several different components of a computer system to process the same data.
What is the revolutionary nature of SHMT? Unlike existing approaches, this method combines the computing power of the CPU, GPU, tensor kernels, and other components. Each of them works in parallel on the same data.
According to a study conducted by scientists, SHMT can speed up data processing by almost 2 times compared to using only a GPU. At the same time, energy consumption is reduced by 51%. These results were achieved without adding new components to the computer system – simply by using the existing ones more efficiently.
In their experiments, the authors of the discovery, Hung-Wei Tseng and Kuan-Chee Xu, managed to make a CPU, Nvidia graphics card, and AI processor work together on the same system platform. According to the developers, modern computers do not fully utilize their potential due to the limitations of traditional approaches to parallel computing. SHMT eliminates these limitations, paving the way for significant productivity gains. Of course, for the practical application of SHMT, improvements in the form of code and hardware optimization are required.
But the potential of this technology is enormous. Its implementation can benefit both large IT companies and ordinary users. In particular, SHMT will accelerate the development of artificial intelligence, reduce the cost and increase the performance of computers, reduce energy consumption by data centers, and improve the performance of mobile devices, among other things. A study by scientists from California related to parallel computing may pave the way for a new approach to computer technology development.
