UpDLRM: Accelerating Personalized Recommendation using Real-World PIM Architecture

TL;DR

UpDLRM leverages real-world PIM hardware to significantly accelerate deep learning recommendation models by enhancing memory bandwidth and reducing inference latency, outperforming CPU and GPU solutions.

Contribution

The paper introduces UpDLRM, a novel approach that utilizes PIM hardware to optimize embedding lookup performance in DLRMs, addressing memory bottlenecks.

Findings

UpDLRM achieves lower inference times compared to CPU-only systems.

Utilizes PIM hardware to improve memory bandwidth for recommendation models.

Effective embedding table partitioning enhances workload balance and caching.

Abstract

Deep Learning Recommendation Models (DLRMs) have gained popularity in recommendation systems due to their effectiveness in handling large-scale recommendation tasks. The embedding layers of DLRMs have become the performance bottleneck due to their intensive needs on memory capacity and memory bandwidth. In this paper, we propose UpDLRM, which utilizes real-world processingin-memory (PIM) hardware, UPMEM DPU, to boost the memory bandwidth and reduce recommendation latency. The parallel nature of the DPU memory can provide high aggregated bandwidth for the large number of irregular memory accesses in embedding lookups, thus offering great potential to reduce the inference latency. To fully utilize the DPU memory bandwidth, we further studied the embedding table partitioning problem to achieve good workload-balance and efficient data caching. Evaluations using real-world datasets show that, UpDLRM achieves much lower inference time for DLRM compared to both CPU-only and CPU-GPU hybrid counterparts.