Abstract
Despite the increasing investment in integrated GPUs and next-generation interconnect research, discrete GPUs connected by PCI Express still account for the dominant position of the market, the management of data communication between CPU and GPU continues to evolve. This paper analyze the address translation overhead and migration latency introduced by this paged memory management solution in CPU-GPU heterogeneous systems. Based on the analysis, a new memory management scheme is proposed: paged memory management solution supporting incomplete pages, which can limit both address translation overhead and migration delay. “Incomplete” refers to a page that has only been partially migrated. This new memory management solution modifies the address translation and data migration process with only minor changes in hardware.
Cite
CITATION STYLE
Shen, L., Zhang, S., Yang, Y., & Wang, Z. (2018). GPU memory management solution supporting incomplete pages. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11276 LNCS, pp. 174–178). Springer Verlag. https://doi.org/10.1007/978-3-030-05677-3_20
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
