Max-PIM: Fast and Efficient Max/Min Searching in DRAM
Event Type
Research Manuscript
Virtual Programs
Hosted in Virtual Platform
Near-Memory and In-Memory Computing
Embedded Systems
DescriptionRecently, in-memory-computing is getting popular to solve the “memory-wall” issue for big-data processing. In this work, for the first time, we propose a novel parallel Min/Max-in-memory algorithm, which supports parallel searching for minimum/maximum of data in DRAM. Our architecture (Max-PIM) supports all basic bit-wise Booleanlogic but optimized for data-transpose and fast one cycle XNOR-logic-in-DRAM operation. Extensive experiments of applying Max-PIM in big-data sorting and graph processing show that it could speed up ̃50X and ̃1000X than GPU and CPU, while only consuming 10% and 1% energy. Comparing with other representative In-DRAM-computing platforms, our design could speed up ̃3X - 10X