BLUE-CHIP: Energy Reduction in Simultaneous Multi-Threaded Processors via Dynamic Resizing of Datapath Resources

Funded Research Project (TÜBİTAK, No: 107E196)
Year: 2010 Project Team: Yrd. Doç. Dr. Gürhan Küçük (gkucuk [at], Mine Mesta.  

Today, the Simultaneous Multi-Threaded (SMT) processors aim to increase the system throughput by executing the instructions coming from different threads in a single clock cycle. However, for achieving this, they require larger (issue queue, physical register files, etc.) and/or multiple copies (reorder buffer, architectural register files, etc.) of datapath resources compared to standard superscalar processors. As a consequence of these modifications, increasing size of die areas and energy consumption are inevitable, in these processors. Nowadays, all the systems, from high-performance servers to battery-operated laptop computers, aim reliability, high-performance and longevity. As it is directly related with these aims, the issue of energy savings in processors is becoming more important day by day.

In this project, the method, which is proposed for SMT processors, aims the resizing of datapath resources (issue queue, physical register files, load/store queue and reorder buffers) (Blue-Chip 1, BC1) and level two (L2) cache (Blue-Chip 2, BC2) according to the demands of the running applications. To achieve this, the targeted resources are divided into several partitions, and turned on and off according to the needs of the application. Since, the energy consumption of the turned-off datapath resources is quite low; as a result, it becomes possible to have great amount of energy savings within a processor. Here, the sharing possibility of some of the structures, which is utilized by only a single-thread in some architectures, by two or four threads, generates new design requirements to the existing method by targeting the fair use of datapath resources by multiple threads.

The energy savings achieved in this project improve the reliability, life time, ergonomic factors and performance characteristics of the simultaneous multithreaded processors, making them feasible even within battery-operated embedded systems.

Permanent link to this article: