This page explains differences, use cases, and how to choose the right resource pool based on customer's needs.
Resource Pools are logical groupings of compute and GPU resources within Denvr’s infrastructure. They determine how workloads are scheduled and what level of availability and performance guarantees apply. There are two main types of resource pools:
On-Demand Resource Pools
Reserved Resource Pools
Each is designed to serve different workload types and customer requirements.
An on-demand resource pool is a shared pool of compute and GPU resources available on a best-effort basis. VMs scheduled in this pool are provisioned only when the requested configuration (e.g., specific GPU, memory, CPU) becomes available. This model is ideal for flexible, cost-effective usage, where resource availability is not time-critical.
Shared Access: The pool is shared across multiple tenants and workloads.
Dynamic Allocation: Resources are allocated only if available at the time of the request.
No Availability Guarantee: Resource allocation is not guaranteed. VMs may remain in a pending state if the required configuration is not currently available.
First-Come, First-Served: Scheduling is non-prioritized and based solely on real-time availability.
Development or testing environments
Batch or burst workloads with no strict SLA
Non-critical jobs that can tolerate delays
VMs may become unschedulable during peak usage or when demand for certain configurations (e.g., A100 80GB GPUs) exceeds current capacity.
Users may see errors such as 0/45 nodes are available: 45 Insufficient nvidia.com/gpu on our dashboard when creating VMs.
A reserved resource pool is a dedicated allocation of compute and GPU resources set aside specifically for a customer or workload. These resources are provisioned and guaranteed in advance, providing predictable availability and performance.
Guaranteed Availability: Resources are reserved and pre-allocated to your organization.
Single-Tenant or Shared-Internal Use: Pools can be configured for single-customer isolation or shared use across internal workloads.
Consistent Performance: Reserved capacity ensures consistent scheduling and uptime, especially for production workloads.
Custom Configuration: Customers can specify GPU type, memory, CPU, and network requirements tailored to their needs.
Production workloads with strict SLA requirements
Workloads requiring consistent and uninterrupted access to specific hardware configurations
Multi-VM environments with dependencies between nodes
AI/ML training pipelines requiring large contiguous GPU allocations
To set up a reserved pool:
Contact your Account Manager or our Customer Experience team.
We will assist in:
Evaluating your workload profile
Designing and sizing the reserved pool
Provisioning dedicated infrastructure
Access controls and tenancy setup will be handled as part of the onboarding.
Last updated
