Increasing computer processing while decreasing energy consumption. What could be better?
And indeed, that’s the outcome of research at the University of California, Riverside.
The key is a new framework, called simultaneous and heterogeneous multithreading (SHMT), a type of parallel processing in which the computational function of multiple components is broken up and shared.
SHMT’s runtime system separates virtual operations into one or more high-level operations (HLOPs) to take advantage of multiple hardware resources simultaneously, allocating HLOPs to the target hardware’s task queues and adjusting task assignments as necessary.
In tests, SHMT was found to be 1.95 times faster and use 51% less energy than the baseline method.
Hung-Wei Tseng, a UC-Riverside associate professor of electrical and computer engineering and co-lead author on the study, commented that the implications for SHMT are huge. Phones, tablets, desktops, laptops could achieve performance gains while reducing the need for expensive, high-performance components, leading to cheaper and more efficient devices.
And who doesn’t want a more efficient way to play solitaire on their phone?