Serpens: A High Bandwidth Memory Based Accelerator for General-Purpose Sparse Matrix-Vector Multiplication
TimeTuesday, July 12th1:30pm - 1:53pm PDT
Location3002, Level 3
Event Type
Research Manuscript
AI/ML Design: Circuits and Architecture
DescriptionSparse matrix-vector multiplication (SpMV) plays a crucial role in many applications. High bandwidth memory (HBM) based FPGAs are a good fit for SpMV accelerators. In this paper, we presentSerpens, an HBM based accelerator for general-purpose SpMV. Serpens features (1) a general-purpose design, (2) memory-centric processing engines, and (3) index coalescing. Evaluations show that Serpens is 1.91x and 1.76x better in geomean throughput than the latest accelerators GraphLiLy and Sextans, respectively, and achieves 2.10x higher throughput than a K80 GPU. After scaling up to 24 HBM channels, Serpens achieves up to 30,204MTEPS and up to 3.79x over GraphLily.