JCSE, vol. 12, no. 3, pp.106-114, 2018

DOI: http://dx.doi.org/10.5626/JCSE.2018.12.3.106

Exploring GPU Data Cache Leakage Management Techniques

Hao Wen and Wei Zhang
Department of Electrical and Computer Engineering, Virginia Commonwealth University, Richmond, VA, USA

Abstract: In this paper, we study how to reduce cache leakage energy efficiently for GPU data caches (L1 and L2). The access pattern of GPU cache is different from that of the CPU, which usually has little locality and high miss rate. In addition, GPU can hide memory latency more effectively due to multi-threading. Because of the above reasons, we find it is possible to place cache lines of GPU data caches into the low power mode more aggressively than traditional leakage management for CPU caches that can reduce more energy leakage without significant performance degradation. In some cases, we find it is possible for the GPU to bypass the L1 data cache to save 100% energy leakage while generating better performance. Also, we propose to combine the drowsy and gated-VDD techniques that can exploit short and long access intervals to minimize energy leakage with insignificant performance overhead. Interestingly, we may achieve better performance for some benchmarks due to the different cache access patterns after applying the leakage reduction method.

Keyword: GPU; Cache; leakage reduction; Two-level low power mode

Full Paper:   316 Downloads, 1375 View