LogoLogo
Contact support
v2.21
  • Home
  • SaaS
  • Self-hosted
v2.21
  • Getting Started
    • Overview
    • What's New
      • What’s New in Version 2.21
      • Hotfixes for Version 2.21
    • Installation
      • Control Plane System Requirements
      • Preparations
      • Network Requirements
      • Install the Control Plane
      • Cluster System Requirements
      • Install Using Helm
      • Customized Installation
      • Upgrade
      • Uninstall
  • Infrastructure setup
    • Authentication and Authorization
      • Authentication and Authorization
      • Users
      • SSO
        • Set Up SSO with SAML
        • Set Up SSO with OpenID Connect
        • Set Up SSO with OpenShift
      • Roles
      • Applications
      • User Applications
      • Access Rules
    • Advanced Setup
      • Node Roles
      • Advanced Control Plane Configurations
      • Advanced Cluster Configurations
      • Service Mesh
      • Integrations
        • Interworking with Karpenter
    • Infrastructure Procedures
      • NVIDIA Run:ai at Scale
      • Monitoring and Maintenance
      • NVIDIA Run:ai System Monitoring
      • Clusters
      • Shared Storage
      • Nodes Maintenance
      • Cluster Restore
      • Secure Your Cluster
      • Logs Collection
      • Event History
  • Platform management
    • Manage AI Initiatives
      • Adapting AI Initiatives to Your Organization
      • Managing Your Organization
        • Projects
        • Departments
      • Managing Your Resources
        • Nodes
        • Configuring NVIDIA MIG Profiles
        • Using GB200 NVL72 and Multi-Node NVLink Domains
        • Node Pools
    • Scheduling and Resource Optimization
      • Scheduling
        • The NVIDIA Run:ai Scheduler: Concepts and Principles
        • How the Scheduler Works
        • Workload Priority Control
        • Quick Starts
          • Over Quota, Fairness and Preemption
      • Resource Optimization
        • GPU Fractions
        • Dynamic GPU Fractions
        • Optimize Performance with Node Level Scheduler
        • GPU Time-Slicing
        • GPU Memory Swap
        • Quick Starts
          • Launching Workloads with GPU Fractions
          • Launching Workloads with Dynamic GPU Fractions
          • Launching Workloads with GPU Memory Swap
    • Policies
      • Policies and Rules
      • Workload Policies
      • Policy YAML Examples
      • Policy YAML Reference
      • Scheduling Rules
    • Monitor Performance and Health
      • Before You Start
      • Metrics and Telemetry
      • Reports
  • Workloads in NVIDIA Run:ai
    • Introduction to Workloads
    • NVIDIA Run:ai Workload Types
    • Workloads
    • Workload Assets
      • Workload Assets
      • Environments
      • Data Sources
      • Data Volumes
      • Compute Resources
      • Credentials
    • Workload Templates
      • Workspace Templates
    • Experiment Using Workspaces
      • Running Workspaces
      • Quick Starts
        • Running Jupyter Notebooks Using Workspaces
    • Train Models Using Training
      • Standard Training
        • Train Models Using a Standard Training Workload
        • Quick Starts
          • Run Your First Standard Training
      • Distributed Training
        • Train Models Using a Distributed Training Workload
        • Quick Starts
          • Run Your First Distributed Training
    • Deploy Models Using Inference
      • Deploy a Custom Inference Workload
      • Deploy Inference Workloads from Hugging Face
      • Deploy Inference Workloads with NVIDIA NIM
      • Deploy NVIDIA Cloud Functions (NVCF) in NVIDIA Run:ai
  • Reference
    • CLI Reference
      • Install and Configure CLI
      • Administrator CLI
      • Add NVIDIA Run:ai Authorization to Kubeconfig
      • CLI Commands Reference
        • runai cluster
        • runai config
        • runai inference
        • runai jax
        • runai kubeconfig
        • runai login
        • runai logout
        • runai mpi
        • runai node
        • runai nodepool
        • runai project
        • runai pvc
        • runai pytorch
        • runai report
        • runai tensorflow
        • runai training
        • runai upgrade
        • runai version
        • runai whoami
        • runai workload
        • runai workspace
        • runai xgboost
      • CLI Commands Examples
    • API Reference
      • How to Authenticate to the API
      • NVIDIA Run:ai REST API
        • Configuring Slack Notifications
  • Support Policy
    • Product Support Policy
    • Product Version Life Cycle
On this page
Export as PDF
  1. Platform management
  2. Scheduling and Resource Optimization
  3. Resource Optimization

Quick Starts

Launching Workloads with GPU FractionsLaunching Workloads with Dynamic GPU FractionsLaunching Workloads with GPU Memory Swap
PreviousGPU Memory SwapNextLaunching Workloads with GPU Fractions

Last updated 2 months ago

LogoLogo

Corporate Info

  • NVIDIA.com Home
  • About NVIDIA
  • Privacy Policy
  • Manage My Privacy
  • Terms of Service

NVIDIA Developer

  • Developer Home
  • Blog

Resources

  • Contact Us
  • Developer Program

Copyright © 2025, NVIDIA Corporation.