Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough

Javier Prades; Carlos Reaño; Federico Silla

doi:10.1016/j.future.2025.107944

Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough

Javier Prades^*
, Carlos Reaño
, Federico Silla

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Virtualization has been the key element for the growth of cloud computing. Historically, GPUs have been complex devices to virtualize efficiently. The mediated passthrough mechanism is usually leveraged. However, it implies a rigid association between the virtual domain and the virtual GPU, which impairs overall system GPU performance. In this paper we propose the use of Network GPGPU system (NGS) to improve the performance of GPUs in virtual domains. Our proposal is compared to NVIDIA vGPU, the most widely used mechanism for virtualizing CUDA-enabled GPUs today. Results show throughput benefits of approximately 20%, speedup of up to 28% in the execution time of the applications, improved overall GPU utilization (over 85%), and lower energy consumption per job (up to 15.34%).

Original language	English
Article number	107944
Number of pages	14
Journal	Future Generation Computer Systems
Volume	174
Early online date	18 Jun 2025
DOIs	https://doi.org/10.1016/j.future.2025.107944
Publication status	Published - Jan 2026
Externally published	Yes

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

CUDA
GPU acceleration
Mediated device passthrough
NGS
NVIDIA vGPU
rCUDA
Virtualization

ASJC Scopus subject areas

Software
Hardware and Architecture
Computer Networks and Communications

Access to Document

10.1016/j.future.2025.107944Licence: CC BY-NC-ND

Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough
Copyright 2025 the authors. This is an open access article published under a Creative Commons Attribution-NonCommercial-NoDerivs License (https://creativecommons.org/licenses/by-nc-nd/4.0/), which permits distribution and reproduction for non-commercial purposes, provided the author and source are cited.
Final published version, 1.69 MBLicence: CC BY-NC-ND

Cite this

@article{31b28264951844e3bb7bbd9e6074f7fc,

title = "Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough",

abstract = "Virtualization has been the key element for the growth of cloud computing. Historically, GPUs have been complex devices to virtualize efficiently. The mediated passthrough mechanism is usually leveraged. However, it implies a rigid association between the virtual domain and the virtual GPU, which impairs overall system GPU performance. In this paper we propose the use of Network GPGPU system (NGS) to improve the performance of GPUs in virtual domains. Our proposal is compared to NVIDIA vGPU, the most widely used mechanism for virtualizing CUDA-enabled GPUs today. Results show throughput benefits of approximately 20\%, speedup of up to 28\% in the execution time of the applications, improved overall GPU utilization (over 85\%), and lower energy consumption per job (up to 15.34\%).",

keywords = "CUDA, GPU acceleration, Mediated device passthrough, NGS, NVIDIA vGPU, rCUDA, Virtualization",

author = "Javier Prades and Carlos Rea{\~n}o and Federico Silla",

year = "2026",

month = jan,

doi = "10.1016/j.future.2025.107944",

language = "English",

volume = "174",

journal = "Future Generation Computer Systems",

issn = "0167-739X",

publisher = "Elsevier B.V.",

}

TY - JOUR

T1 - Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough

AU - Prades, Javier

AU - Reaño, Carlos

AU - Silla, Federico

PY - 2026/1

Y1 - 2026/1

N2 - Virtualization has been the key element for the growth of cloud computing. Historically, GPUs have been complex devices to virtualize efficiently. The mediated passthrough mechanism is usually leveraged. However, it implies a rigid association between the virtual domain and the virtual GPU, which impairs overall system GPU performance. In this paper we propose the use of Network GPGPU system (NGS) to improve the performance of GPUs in virtual domains. Our proposal is compared to NVIDIA vGPU, the most widely used mechanism for virtualizing CUDA-enabled GPUs today. Results show throughput benefits of approximately 20%, speedup of up to 28% in the execution time of the applications, improved overall GPU utilization (over 85%), and lower energy consumption per job (up to 15.34%).

AB - Virtualization has been the key element for the growth of cloud computing. Historically, GPUs have been complex devices to virtualize efficiently. The mediated passthrough mechanism is usually leveraged. However, it implies a rigid association between the virtual domain and the virtual GPU, which impairs overall system GPU performance. In this paper we propose the use of Network GPGPU system (NGS) to improve the performance of GPUs in virtual domains. Our proposal is compared to NVIDIA vGPU, the most widely used mechanism for virtualizing CUDA-enabled GPUs today. Results show throughput benefits of approximately 20%, speedup of up to 28% in the execution time of the applications, improved overall GPU utilization (over 85%), and lower energy consumption per job (up to 15.34%).

KW - CUDA

KW - GPU acceleration

KW - Mediated device passthrough

KW - NGS

KW - NVIDIA vGPU

KW - rCUDA

KW - Virtualization

U2 - 10.1016/j.future.2025.107944

DO - 10.1016/j.future.2025.107944

M3 - Article

AN - SCOPUS:105008296027

SN - 0167-739X

VL - 174

JO - Future Generation Computer Systems

JF - Future Generation Computer Systems

M1 - 107944

ER -

Enhancing the performance of GPU acceleration in virtual environments: Thoroughly benchmarking the rigidity of mediated device passthrough

Abstract

UN SDGs

Keywords

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this