Skip to main content
GPU driver management in CKS has the following limitations:
  • No Cloud Console support: You must configure GPU driver management through Kubernetes manifests.
  • Limited version updates: You can only select major version updates. When minor version updates are available, CKS generates a new pending configuration on the Node Pool. Find pending configurations on the Node Pool’s status.pendingNodeConfiguration. See Manage Node Pool Configuration for more details.
  • No release channel support: Node Pools do not support the latest and stable release channels.
Contact CoreWeave support if you have questions about GPU driver management.

​
Do not install the NVIDIA GPU Operator on CKS clusters

CoreWeave manages the NVIDIA GPU Operator on your behalf. Do not install the NVIDIA GPU Operator on CKS clusters. Doing so conflicts with the platform-managed deployment and is not supported.
This page explains how GPU driver management works in CoreWeave Kubernetes Service (CKS) so you can pin your Node Pools to a supported driver version and route workloads to Nodes that match. Use this approach when your training, inference, or high-performance computing (HPC) workloads depend on a specific NVIDIA driver version for compatibility, performance, or reproducibility. CKS lets you specify a GPU driver version for your Node Pools. You can specify driver versions in the Node Pool manifest when deploying a Node Pool. CKS workloads can target Nodes running specific driver versions using the nodeSelector field in the workload manifest.
Selecting a driver version using the gpu field is optional. If you don’t specify a version, the Node Pool uses the latest available driver version.

​
Supported driver versions

Use the following table to confirm which driver version to pin when configuring a Node Pool. CoreWeave supports all tested release driver (TRD) versions from NVIDIA, listed in the following table as major versions.
Although you aren’t required to upgrade from a major version, CoreWeave recommends that you upgrade frequently to ensure the best combination of testing coverage, security posture, stability, and performance.
NameInstance IDDefault GPU driverCompatible GPU driver
GB300 NVL72 Quantum-X InfiniBandgb300-4x580580, 595
GB300 NVL72 Spectrum-X RoCEgb300-4x-e580580, 595
GB200 NVL72-powered instancesgb200-4x595580, 595
B300 (InfiniBand)b300-8x580580, 595
B200 (InfiniBand)b200-8x595580, 595
H200 (InfiniBand)gd-8xh200ib-i128595535, 580, 595
H100 (InfiniBand)gd-8xh100ib-i128595535, 580, 595
RTX Pro 6000 Blackwell Server Editionrtxp6000-8x580535, 580, 595
L40Sgd-8xl40s-i128595535, 580, 595
L40gd-8xl40-i128595535, 580, 595
GH200gd-1xgh200595580, 595
A100gd-8xa100-i128580535, 580, 595

​
Additional resources

For step-by-step procedures and related reference material, see:
Last modified on June 10, 2026