Platform Integration
GPU-as-a-Service
Offer GPU-accelerated workloads for AI/ML, VDI, and HPC with native GPU support and usage-based billing.
Monetize GPU Workloads with StackBill
Turn your GPU infrastructure into a revenue-generating service with complete billing integration
Self-Service GPU Portal
Enable your customers to provision GPU-backed instances on-demand through a complete self-service portal with real-time availability.
Learn moreGPU Usage-Based Billing
Monetize GPU workloads with granular metering. Track GPU allocation by time, type, and tenant with automatic chargeback and invoicing.
Learn moreGPU Service Offerings
Create differentiated GPU compute offerings - from entry-level inference to high-end training clusters - each with custom pricing.
Learn moreResource Limits & Quotas
Enforce GPU limits at Account, Domain, and Project levels. Prevent resource hoarding and ensure fair allocation across tenants.
Learn moreGPU Utilization Analytics
Monitor GPU usage across your infrastructure. Identify underutilized resources and optimize placement for maximum efficiency.
Learn moreMulti-Tenant Isolation
Complete tenant isolation for GPU resources. Each customer's GPU workloads are securely separated at the hardware level.
Learn moreEnterprise-Grade GPU Features
Native GPU orchestration powered by Apache CloudStack 4.21 with StackBill monetization
Supported GPU Vendors & Models
CloudStack 4.21 supports major GPU vendors for passthrough and vGPU deployments
Supported GPU Models
Industry-leading GPUs for AI, ML, and professional graphics workloads
Data Center
Professional
Blackwell
Legacy
Supported Features
Deployment Modes
Passthrough
Dedicated GPU
vGPU
Shared GPU
Supported GPU Models
High-performance GPUs for compute and visualization workloads
Instinct
Radeon Pro
Supported Features
Deployment Modes
Passthrough
Dedicated GPU
vGPU
Shared GPU
Supported GPU Models
Flexible GPU solutions for edge AI and media processing
Data Center GPU
Arc Pro
Supported Features
Deployment Modes
Passthrough
Dedicated GPU
vGPU
Shared GPU
Passthrough vs vGPU
Choose the right GPU deployment mode based on your workload requirements
GPU Passthrough
Dedicated GPU assignment where the entire physical GPU is exclusively assigned to a single instance for maximum performance.
- Full GPU resources to single VM
- Maximum performance for AI training
- Ideal for HPC and rendering
- Direct hardware access
vGPU Partitioning
Single physical GPU partitioned for use by multiple instances, enabling efficient resource sharing and higher density.
- Share GPU across multiple VMs
- Higher density, lower cost per VM
- Ideal for VDI and inference
- NVIDIA vGPU software support
Who Benefits from GPU-as-a-Service?
Enable diverse GPU workloads for different customer segments
AI/ML Training & Inference
Offer GPU compute for machine learning model training and real-time inference workloads.
Learn moreVirtual Desktop (VDI)
Deliver GPU-accelerated virtual desktops for CAD, design, and professional applications.
Learn moreRendering & Visualization
Enable render farms and 3D visualization workloads for media and entertainment customers.
Learn moreScientific Computing (HPC)
Support high-performance computing for research, simulations, and data analytics.
Learn moreFrequently Asked Questions
CloudStack 4.21 supports two GPU deployment modes: Passthrough (dedicated) where a GPU is exclusively assigned to a single instance, and vGPU (partitioned) where a single physical GPU is shared across multiple instances. The mode depends on your GPU hardware capabilities and use case requirements.
CloudStack automatically discovers GPU devices on supported KVM hosts. Administrators can then group and classify GPUs by vendor or type, create GPU-backed Service Offerings, and set quota limits at Account, Domain, and Project levels. End users select GPU offerings when deploying instances.
Yes, GPU-backed Service Offerings can be used with CloudStack Kubernetes Service (CKS). This enables customers to deploy GPU-accelerated Kubernetes clusters for machine learning training, inference, and other GPU-intensive containerized workloads.
StackBill provides granular GPU usage tracking and billing. GPU allocation is metered throughout the full lifecycle - from provisioning to termination. Custom pricing can be set per GPU type, and usage is automatically included in tenant invoices with detailed breakdowns.
CloudStack supports NVIDIA vGPU software for partitioning GPUs across multiple VMs. Supported GPUs include L-series (L2, L4, L20, L40, L40S), A-series (A2, A10, A16, A40), T4, V100, and professional RTX cards. Note: C-series vGPU types require NVIDIA AI Enterprise licensing.
Absolutely. GPU resources are fully integrated with CloudStack's multi-tenant architecture. Each tenant's GPU allocations are isolated, quota limits are enforced per Account/Domain/Project, and usage is tracked separately for billing and chargeback purposes.
Ready to Offer GPU-as-a-Service?
Enable AI/ML, VDI, and HPC workloads for your customers with CloudStack's native GPU support and StackBill's monetization capabilities.