- Overview
- Requirements
- Pre-installation
- Preparing the installation
- Downloading the installation packages
- Configuring the OCI-compliant registry
- Granting installation permissions with cluster roles
- Granting installation permissions without cluster roles
- Installing and configuring the service mesh
- Installing and configuring the GitOps tool
- Installing the External Secrets Operator
- Deploying Redis through OperatorHub
- Applying miscellaneous configurations
- Running uipathctl
- Installation
- Post-installation
- Migration and upgrade
- Monitoring and alerting
- Cluster administration
- Product-specific configuration
- Orchestrator advanced configuration
- Configuring Orchestrator parameters
- Configuring appSettings
- Configuring the maximum request size
- Overriding cluster-level storage configuration
- Configuring NLog
- Saving robot logs to Elasticsearch
- Configuring credential stores
- Configuring encryption key per tenant
- Cleaning up the Orchestrator database
- Skipping host library installation
- AI Trust Layer
- Troubleshooting
- How to uninstall Automation Suite
- How to resolve prereq connectivity check failure on OpenShift 4.16–4.18
- How to deploy Insights in a FIPS-enabled cluster
- How to disable CDI auto-enabling in Nvidia GPU Operator
- How to URL-encode a PostgreSQL connection string password
- 500 errors and rate limiting on S3 requests in ODF
- External Secrets troubleshooting
- Configuring CSI driver tolerations for ODF 4.20
- Temporal as a Service troubleshooting
- AI Center and Document Understanding pods fail to start with TLS certificate verification enabled
- Studio Desktop cannot load Integration Service connectors and activities
- TLS certificate validation errors
- Manual ArgoCD NetworkPolicy mitigation (GHSA-47m3-95c7-g2g8)
- Configuring resource requests and limits for uipathctl-created workloads
How to disable CDI auto-enabling in Nvidia GPU Operator
Disable Container Device Interface in Nvidia GPU Operator for Automation Suite GPU workloads.
The NVIDIA GPU Operator enables Container Device Interface (CDI) by default in recent versions. In earlier versions of the operator, CDI was disabled by default.
When CDI is enabled, the operator adds an nvidia-cdi-hook. This hook requires all pod specifications that target GPU nodes to explicitly set the following field:
runtimeClassName: nvidia
runtimeClassName: nvidia
Automation Suite workloads do not include this setting. As a result, GPU workloads fail to start.
To prevent this issue, you must disable CDI in the gpu-cluster-policy ClusterPolicy by setting enabled to false:
spec:
cdi:
enabled: false
spec:
cdi:
enabled: false
After you apply this change, CDI is disabled and Automation Suite GPU workloads run without requiring the runtimeClassName setting.