[rheindt20arcs] | Sven Rheindt, Andreas Fried, Oliver Lenke, Lars Nolte, Tim Twardzik, Thomas Wild, Andreas Herkersdorf, X-CEL: A Method to Estimate Near-Memory Acceleration Potential in Tile-based MPSoCs, Proceedings of the 33rd International Conference on Architecture of Computing Systems, Springer, 2020.
|
Abstract
Near-memory acceleration strives to tackle the data-to-task
locality issue in MPSoCs in order to obtain higher performance and lower
power consumption. However, it is not easy to determine whether the
advantages arise from the near-memory integration or the hardware ac-
celeration (versus software execution). We propose X-CEL, a method to
accurately estimate the potential of near-memory acceleration using an
easy-to-integrate near-memory core. We showcase X-CEL's benefits with
three variants of graph copy mechanisms in a tile-based MPSoC. Evalu-
ations reveal that the estimated speedup is in good accordance with the
actual speedup achieved by the near-memory accelerator.
Download
Original article available at springerlink.com.
BibTeX
Authors at the institute
Projects