high performance computing on graphics processing units: hgpu.org

Matrix multiplication consumes a large fraction of the time taken in many machine-learning algorithms. Thus, accelerator chips that perform matrix multiplication faster than conventional processors or even GPU’s are of increasing interest. In this paper, we demonstrate a method of performing matrix multiplication without a scalar multiplier circuit. In many cases of practical interest, only a single addition and a single on-chip copy operation are needed to replace a multiplication. It thus becomes possible to design a matrix-multiplier chip that, because it does not need time, space- and energy-consuming multiplier circuits, can hold many more processors, and thus provide a net speedup.