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 acceleration (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. Evaluations reveal that the estimated speedup is in good accordance with the actual speedup achieved by the near-memory accelerator.
CITATION STYLE
Rheindt, S., Fried, A., Lenke, O., Nolte, L., Sabirov, T., Twardzik, T., … Herkersdorf, A. (2020). X-cel: A method to estimate near-memory acceleration potential in tile-based mpsocs. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 12155 LNCS, pp. 109–123). Springer. https://doi.org/10.1007/978-3-030-52794-5_9
Mendeley helps you to discover research relevant for your work.