A comprehensive evaluation of spatial co-execution on GPUs using MPS and MIG technologies
Preprint
- 11 November 2025
- preprint
- Published by Springer Nature in Research Square
Abstract
To mitigate the increasingly common underutilization of computational resources in modern GPUs, spatial sharing methods enable multiple applications to use them simultaneously. This work presents a comprehensive evaluation of NVIDIA's primary technologies to achieve that goal: Multi-Process Service (MPS) and Multi-Instance GPU (MIG). Our findings reveal a crucial trade-off between MPS's flexibility and MIG's isolation, and reveals many key insights for improving the co-execution strategy according to jobs profiles. In the most favorable scenarios, MPS improves performance by up to 30% and reduces energy by about 20%, using its provisioning option to avoid resource monopolization. However, under memory contention, it suffers severe degradation, worsening performance by around 30%. Conversely, MIG's full hardware isolation resolves memory contention, leading to more consistent improvements, but these gains are tempered by higher overhead, and its rigid scheme can degrade performance in certain cases.Keywords
All Related Versions
This publication has 19 references indexed in Scilit:
- Solving the task scheduling and GPU reconfiguration problem on MIG devices via deep reinforcement learningFuture Generation Computer Systems, 2025
- Leveraging Multi-Instance GPUs through moldable task schedulingJournal of Parallel and Distributed Computing, 2025
- Evaluation of Juliana Tool: A translator for Julia’s CUDA.jl code into KernelAbstraction.jlFuture Generation Computer Systems, 2025
- Granularity- and Interference-Aware GPU Sharing with MPS2024
- Exploring the Diversity of Multiple Job Deployments over GPUs for Efficient Resource Sharing2024
- Top-Down Performance Profiling on NVIDIA's GPUs2022
- The transformational role of GPU computing and deep learning in drug discoveryNature Machine Intelligence, 2022
- A Survey of GPU Multitasking Methods Supported by Hardware ArchitectureIEEE Transactions on Parallel and Distributed Systems, 2021
- Altis: Modernizing GPGPU Benchmarks2020
- Rodinia: A benchmark suite for heterogeneous computing2009