August 15, 2025 - GPU driver management features now available
GPU driver management features are now available in both CKS and SUNK, allowing users to target specific driver versions for their workloads
Update CKS GPU driver management features are now available in CKS Node Pools, allowing you to specify and target specific GPU driver versions for your workloads. This feature provides better control over driver compatibility and enables homogeneous driver environments across your clusters.
Overview
GPU driver management is currently a preview feature with the following limitations:
- No Cloud Console UI: Configuration must be done through Kubernetes manifests or SLURM configuration
- Limited to major versions: Only major driver version selection is currently supported
- Manual upgrade process: Driver version updates require manual configuration changes
GPU driver management features are now available across CoreWeave's compute platforms, providing enhanced control over driver versions and improved workload compatibility. These features allow you to:
- Target specific driver versions in both CKS Node Pools and SUNK SLURM workloads
- Ensure homogeneous driver environments across your compute resources
- Maintain compatibility with specific CUDA versions and application requirements
- Plan driver upgrades with better visibility and control
Supported driver versions
CoreWeave supports all tested release driver (TRD) versions from NVIDIA. See Supported driver versions for more information.
Key benefits
| Benefit | Description |
|---|---|
| Driver compatibility | Ensure workloads run on compatible driver versions |
| Homogeneous environments | Maintain consistent driver versions across compute resources |
| Upgrade planning | Better visibility and control over driver upgrade schedules |
| Application stability | Reduce driver-related compatibility issues in production workloads |
| Resource optimization | Target workloads to Nodes with optimal driver versions for their needs |
Usage
CKS
- Node Pool level configuration: Configure Node Pools to use a specific driver version using the
gpu.versionfield in the Node Pool manifest - Pod scheduling: Schedule Pods onto Nodes specified driver versions using Node selectors and affinity rules
- Kubernetes-native integration: Use Node selectors and affinity rules to schedule Pods onto Nodes with specific driver versions
- Helm chart support: Use the CoreWeave Helm chart to deploy and manage GPU driver versions in your CKS clusters
See Update GPU driver version in CKS Node Pools for more detailed information.