SiPipe: Bridging the CPU-GPU Utilization Gap for Efficient Pipeline-Parallel LLM Inference

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: He, Yongchao, Zhao, Bohan, Cao, Zheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912453532581888
author He, Yongchao
Zhao, Bohan
Cao, Zheng
author_facet He, Yongchao
Zhao, Bohan
Cao, Zheng
contents As inference workloads for large language models (LLMs) scale to meet growing user demand, pipeline parallelism (PP) has become a widely adopted strategy for multi-GPU deployment, particularly in cross-node setups, to improve key-value (KV) cache capacity and inference throughput. However, PP suffers from inherent inefficiencies caused by three types of execution bubbles-load-imbalance, intra-stage, and inter-stage-which limit pipeline saturation. We present SiPipe, a heterogeneous pipeline design that improves throughput by leveraging underutilized CPU resources to offload auxiliary computation and communication. SiPipe incorporates three key techniques-CPU sampling, a token-safe execution model, and structure-aware transmission-to mitigate pipeline bubbles and improve execution efficiency. Across diverse LLMs, SiPipe achieves up to 2.1 times higher throughput, 43% lower per-token latency, and up to 23% higher average GPU utilization compared to the state-of-the-art vLLM under the same PP configuration, demonstrating its generality across LLMs and deployment scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2506_22033
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SiPipe: Bridging the CPU-GPU Utilization Gap for Efficient Pipeline-Parallel LLM Inference
He, Yongchao
Zhao, Bohan
Cao, Zheng
Distributed, Parallel, and Cluster Computing
As inference workloads for large language models (LLMs) scale to meet growing user demand, pipeline parallelism (PP) has become a widely adopted strategy for multi-GPU deployment, particularly in cross-node setups, to improve key-value (KV) cache capacity and inference throughput. However, PP suffers from inherent inefficiencies caused by three types of execution bubbles-load-imbalance, intra-stage, and inter-stage-which limit pipeline saturation. We present SiPipe, a heterogeneous pipeline design that improves throughput by leveraging underutilized CPU resources to offload auxiliary computation and communication. SiPipe incorporates three key techniques-CPU sampling, a token-safe execution model, and structure-aware transmission-to mitigate pipeline bubbles and improve execution efficiency. Across diverse LLMs, SiPipe achieves up to 2.1 times higher throughput, 43% lower per-token latency, and up to 23% higher average GPU utilization compared to the state-of-the-art vLLM under the same PP configuration, demonstrating its generality across LLMs and deployment scenarios.
title SiPipe: Bridging the CPU-GPU Utilization Gap for Efficient Pipeline-Parallel LLM Inference
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2506.22033