In modern computing hardware, the performance gap between processor and memory is one of the most significant factors that limits overall performance improvement of computing system. Also, with the advent of multicore and manycore system, memory bandwidth per core is decreasing constantly. To solve this problem, recently, many researchers are interested in Processing-In-Memory (PIM). PIM is that processing elements are attached to memory-side, so near-memory-processing which is suitable for memory intensive application can be possible. Various researches studied PIM, but it was just single-channel memory system. In addition, PIM is a new architecture that is different with conventional computing system. Thus, common data layout cannot become optimal case for PIM. Optimal data layout is also needed to be studied. In this paper, we propose the multi-channel PIM architecture with PIM-to-PIM communication, because data that is needed to operate can be distributed over several channels. To utilize multichannel PIM architecture properly, we also introduce data layout that can minimize the number of PIM-to-PIM communications which are overheads of the system and maximize parallelism to reduce execution time. We evaluate it about vector arithmetic operation. The result is that execution time is improved about 393% and compared to the worst case, in the optimal data layout.
|Title of host publication||Proceedings of 2018 7th International Conference on Software and Computer Applications, ICSCA 2018|
|Publisher||Association for Computing Machinery|
|Number of pages||5|
|Publication status||Published - 2018 Feb 8|
|Event||7th International Conference on Software and Computer Applications, ICSCA 2018 - Kuantan, Malaysia|
Duration: 2018 Feb 8 → 2018 Feb 10
|Name||ACM International Conference Proceeding Series|
|Other||7th International Conference on Software and Computer Applications, ICSCA 2018|
|Period||18/2/8 → 18/2/10|
Bibliographical notePublisher Copyright:
© 2018 Association for Computing Machinery.
All Science Journal Classification (ASJC) codes
- Human-Computer Interaction
- Computer Vision and Pattern Recognition
- Computer Networks and Communications