Saved in:
Bibliographic Details
Main Authors: Rhee, Myunghyun, Sim, Joonseop, Ahn, Taeyoung, Lee, Seungyong, Yoon, Daegun, Kim, Euiseok, Park, Kyoung, Joo, Youngpyo, Kim, Hoshik
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.16112
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908719106752512
author Rhee, Myunghyun
Sim, Joonseop
Ahn, Taeyoung
Lee, Seungyong
Yoon, Daegun
Kim, Euiseok
Park, Kyoung
Joo, Youngpyo
Kim, Hoshik
author_facet Rhee, Myunghyun
Sim, Joonseop
Ahn, Taeyoung
Lee, Seungyong
Yoon, Daegun
Kim, Euiseok
Park, Kyoung
Joo, Youngpyo
Kim, Hoshik
contents The attention layer, a core component of Transformer-based LLMs, brings out inefficiencies in current GPU systems due to its low operational intensity and the substantial memory requirements of KV caches. We propose a High-bandwidth Processing Unit (HPU), a memoryintensive co-processor that enhances GPU resource utilization during large-batched LLM inference. By offloading memory-bound operations, the HPU allows the GPU to focus on compute-intensive tasks, increasing overall efficiency. Also, the HPU, as an add-on card, scales out to accommodate surging memory demands driven by large batch sizes and extended sequence lengths. In this paper, we show the HPU prototype implemented with PCIe-based FPGA cards mounted on a GPU system. Our novel GPU-HPU heterogeneous system demonstrates up to 4.1x performance gains and 4.6x energy efficiency improvements over a GPUonly system, providing scalability without increasing the number of GPUs.
format Preprint
id arxiv_https___arxiv_org_abs_2504_16112
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HPU: High-Bandwidth Processing Unit for Scalable, Cost-effective LLM Inference via GPU Co-processing
Rhee, Myunghyun
Sim, Joonseop
Ahn, Taeyoung
Lee, Seungyong
Yoon, Daegun
Kim, Euiseok
Park, Kyoung
Joo, Youngpyo
Kim, Hoshik
Hardware Architecture
Artificial Intelligence
Computation and Language
Distributed, Parallel, and Cluster Computing
The attention layer, a core component of Transformer-based LLMs, brings out inefficiencies in current GPU systems due to its low operational intensity and the substantial memory requirements of KV caches. We propose a High-bandwidth Processing Unit (HPU), a memoryintensive co-processor that enhances GPU resource utilization during large-batched LLM inference. By offloading memory-bound operations, the HPU allows the GPU to focus on compute-intensive tasks, increasing overall efficiency. Also, the HPU, as an add-on card, scales out to accommodate surging memory demands driven by large batch sizes and extended sequence lengths. In this paper, we show the HPU prototype implemented with PCIe-based FPGA cards mounted on a GPU system. Our novel GPU-HPU heterogeneous system demonstrates up to 4.1x performance gains and 4.6x energy efficiency improvements over a GPUonly system, providing scalability without increasing the number of GPUs.
title HPU: High-Bandwidth Processing Unit for Scalable, Cost-effective LLM Inference via GPU Co-processing
topic Hardware Architecture
Artificial Intelligence
Computation and Language
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2504.16112