Dark vs. Dim Silicon and Near-Threshold Computing Extended

Authors:Wang, Liang, Department of Computer ScienceUniversity of Virginia Skadron, Kevin, Department of Computer ScienceUniversity of Virginia

Due to limited scaling of supply voltage, power density is expected to grow in future technology nodes. This increasing power density potentially limits the number of transistors switching at full speed in the future. Near-threshold operation can increase the number of simultaneously active cores, at the expense of much lower operating frequency (“dim silicon”). Although promising to increase overall throughput, dim cores suffer from diminishing returns as the number of cores increases. At this point, hardware accelerators become more efficient alternatives. To explore such a broad design space, we have developed a framework called Lumos to analytically quantify the performance limits of many-core, heterogeneous systems operating at near-threshold voltage. Lumos augments Amdahl’s Law with detailed scaling of frequency and power, calibrated by circuit-level simulations using a modified Predictive Technology Model (PTM) and factors in effects of process variations. While our results show that dim cores do indeed boost throughput, even in the presence of process variations, significant benefits are only achieved in applications with very high parallelism or with novel architectures to mitigate variation. A more beneficial and scalable approach is to use accelerators. However, reconfigurable logic that supports a variety of accelerators is more beneficial than a dedicated, fixed-logic accelerator, unless 1) the dedicated kernel has overwhelming coverage across applications (e.g. twice as large as the total of all others), or 2) the speedup of the dedicated accelerator over the reconfigurable equivalent is significant (e.g. 10x-50x).

All rights reserved (no additional license for public reuse)
Source Citation:

Wang, Liang, and Kevin Skadron. "Dark vs. Dim Silicon and Near-Threshold Computing Extended." University of Virginia Dept. of Computer Science Tech Report (2013).

University of Virginia, Department of Computer Science
Published Date: