Heterogeneous Acceleration Pipeline for Recommendation System Training

arXiv:2204.05436v2 Announce Type: replace-cross
Abstract: Recommendation models rely on deep learning networks and large embedding tables, resulting in computationally and memory-intensive processes. These models are typically trained using hybrid CPU-GPU or GPU-only configurations. The hybrid mode combines the GPU's neural network acceleration with the CPUs' memory storage and supply for embedding tables but may incur significant CPU-to-GPU transfer time. In contrast, the GPU-only mode utilizes High Bandwidth Memory (HBM) across multiple GPUs for storing embedding tables. However, this approach is …

abstract arxiv cpu cpus cs.ai cs.ar cs.lg deep learning embedding gpu hybrid memory network networks neural network pipeline processes recommendation storage tables training type