Researchers at the University of California, Riverside have developed an innovative computing process called simultaneous and heterogeneous multithreading (SHMT) that could significantly speed up devices and reduce energy consumption without the need for new hardware. SHMT leverages the multiple processors—such as CPUs, GPUs, and TPUs—already present in modern devices to execute tasks in parallel, thus overcoming bottlenecks caused by data transfer between different processing units.
In tests involving an ARM Cortex-A57 CPU, an Nvidia GPU, and a Google Edge TPU, SHMT enabled code execution to be nearly twice as fast, with a 51% reduction in energy use. This approach challenges current programming models that tend to underutilize the full processing power of heterogeneous computers by only employing the most efficient units for specific tasks.
While promising, the technology is in its infancy, and there are hurdles to address, including the division and reintegration of computing jobs among different processors without compromising performance or precision. The research was presented at the 56th Annual IEEE/ACM International Symposium on Microarchitecture and suggests a future where software optimization could unlock new levels of performance and efficiency in existing devices.
Read more at ScienceAlert…