Unlocking the Full Potential of Your Existing Devices: A Paradigm Shift in Computer Architecture

Thursday - March 7 2024, 00:43 UTC - 8 months ago

Hung-Wei Tseng proposes a paradigm shift in computer architecture with "Simultaneous and Heterogeneous Multithreading" (SHMT), which aims to use all existing processing units simultaneously, eliminating the bottleneck seen in current systems. Tests on an embedded system platform showed impressive results - a 1.96 times speedup and a 51% reduction in energy consumption. While there is still further investigation needed, SHMT shows potential for significant cost and energy savings.

The constant need for faster and more efficient computing has led to the inclusion of specialized processing units such as graphics processing units (GPUs), hardware accelerators for artificial intelligence (AI) and machine learning (ML), and digital signal processing units in our devices. These components, while greatly enhancing the capabilities of our devices, also create a bottleneck in processing by requiring information to be constantly moved between them. But what if there was a way to use all these processing units simultaneously, without adding any new hardware? Hung-Wei Tseng, a UC Riverside associate professor of electrical and computer engineering, has proposed just that in a paper titled, "Simultaneous and Heterogeneous Multithreading." .

Tseng and his team have developed a paradigm shift in computer architecture which they call SHMT. The idea is to utilize all existing processing units simultaneously through a multi-threading approach, eliminating the need for information to be moved between units. This simultaneous and heterogeneous multithreading has been successfully tested on an embedded system platform using a multi-core ARM processor, an NVIDIA GPU, and a Tensor Processing Unit hardware accelerator. The results were impressive - a 1.96 times speedup and a 51% reduction in energy consumption.

The implications of this paradigm shift are huge. Not only could SHMT potentially reduce computer hardware costs, but it could also have a significant impact on the environment. Servers in data processing centers require large amounts of energy to keep them running, resulting in high carbon emissions. By using all processing units simultaneously, energy consumption could be reduced, leading to a decrease in carbon emissions. Additionally, the need for scarce freshwater to cool servers could also be reduced.

However, Tseng's paper cautions that further investigation is needed to fully understand the implications and limitations of SHMT implementation. Questions regarding system implementation, hardware support, code optimization, and the types of applications that would benefit the most still need to be answered. But with the successful results of their tests, the potential for SHMT to unlock the full potential of our existing devices is clear.

Tseng's paper was presented at the 56th Annual IEEE/ACM International Symposium on Microarchitecture, highlighting the importance of this research in the world of computer architecture.