forde-commit-1770160653

Author: blik616287

packs/nvidia-dra-driver-25.8.1/README.md

@@ -1,56 +1,52 @@
 # NVIDIA DRA Driver for GPUs
 
-This pack installs the NVIDIA Dynamic Resource Allocation (DRA) Driver for GPUs, enabling flexible GPU allocation in Kubernetes 1.32+.
+The [NVIDIA DRA Driver](https://github.com/NVIDIA/k8s-dra-driver-gpu) enables Dynamic Resource Allocation (DRA) for GPUs in Kubernetes 1.32+. This pack works with Palette to provide flexible GPU allocation using DeviceClass and ResourceClaim resources, replacing the traditional device plugin approach with a modern, CEL-based device selection mechanism.
 
-## Overview
-
-DRA is a Kubernetes feature that provides flexible request and sharing of hardware resources like GPUs. The NVIDIA DRA Driver replaces the traditional NVIDIA device plugin approach with a more modern, CEL-based device selection mechanism.
 
 ## Prerequisites
 
-- Kubernetes 1.32 or newer (DRA is GA in 1.34+)
-- NVIDIA GPU Operator 25.3.0+ (for driver management and CDI support)
-- CDI enabled in the container runtime (containerd/CRI-O)
-- Node Feature Discovery (NFD) for GPU detection
+- Kubernetes 1.32 or newer (DRA is GA in 1.34+).
+- [NVIDIA GPU Operator](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/index.html) 25.3.0+ for driver management and CDI support.
+- CDI enabled in the container runtime (containerd/CRI-O).
+- [Node Feature Discovery](https://kubernetes-sigs.github.io/node-feature-discovery/) (NFD) for GPU detection.
 
-## Key Features
 
-- **Dynamic GPU Allocation**: Request GPUs using DeviceClass and ResourceClaim resources
-- **CEL-based Selection**: Filter GPUs by attributes using Common Expression Language
-- **GPU Sharing**: Multiple pods can share access to the same GPU
-- **ComputeDomains**: Support for Multi-Node NVLink (MNNVL) on GB200 systems
+## Parameters
 
-## Configuration
+To deploy the NVIDIA DRA Driver, you can configure the following parameters in the pack's YAML.
 
-### Driver Root Path
+| **Name** | **Description** | **Type** | **Default Value** | **Required** |
+|---|---|---|---|---|
+| `nvidiaDriverRoot` | Path to NVIDIA driver installation. Use `/run/nvidia/driver` with GPU Operator, `/` for host-installed drivers. | String | `/run/nvidia/driver` | No |
+| `resources.gpus.enabled` | Enable GPU allocation via DRA. | Boolean | `true` | No |
+| `resources.computeDomains.enabled` | Enable ComputeDomains for Multi-Node NVLink (MNNVL) on GB200 systems. | Boolean | `false` | No |
+| `image.tag` | DRA driver image tag. | String | `v25.8.1` | No |
+| `logVerbosity` | Log verbosity level (0-7, higher = more verbose). | String | `4` | No |
+| `webhook.enabled` | Enable admission webhook for advanced validation. | Boolean | `false` | No |
 
-When using with GPU Operator (recommended):
-```yaml
-nvidiaDriverRoot: /run/nvidia/driver
-```
+Refer to the [NVIDIA DRA Driver Helm chart](https://github.com/NVIDIA/k8s-dra-driver-gpu) for the complete list of configurable parameters.
 
-When drivers are installed directly on host:
-```yaml
-nvidiaDriverRoot: /
-```
 
-### Enable/Disable Resources
+## Upgrade
+
+N/A - This is the initial release of the NVIDIA DRA Driver pack.
 
-```yaml
-resources:
-  gpus:
-    enabled: true        # Enable GPU allocation
-  computeDomains:
-    enabled: false       # Enable for MNNVL systems
-```
 
 ## Usage
 
+To use the NVIDIA DRA Driver pack, first create a new [add-on cluster profile](https://docs.spectrocloud.com/profiles/cluster-profiles/create-cluster-profiles/create-addon-profile/), search for the **NVIDIA DRA Driver for GPUs** pack, and configure the driver root path based on your environment:
+
+```yaml
+charts:
+  nvidia-dra-driver-gpu:
+    nvidiaDriverRoot: /run/nvidia/driver  # Use "/" if drivers installed on host
+```
+
 After installation, the DRA driver creates:
 - A default `DeviceClass` named `gpu.nvidia.com`
 - `ResourceSlice` objects representing available GPUs on each node
 
-### Example ResourceClaimTemplate
+To request a GPU for your workload, create a ResourceClaimTemplate and reference it in your Pod. Click on the **Add Manifest** button to create a new manifest layer with the following content:
 
 ```yaml
 apiVersion: resource.k8s.io/v1
@@ -63,11 +59,7 @@ spec:
       requests:
         - name: gpu
           deviceClassName: gpu.nvidia.com
-```
-
-### Example Pod Using DRA
-
-```yaml
+---
 apiVersion: v1
 kind: Pod
 metadata:
@@ -84,8 +76,12 @@ spec:
       resourceClaimTemplateName: gpu-claim
 ```
 
-## Documentation
+Once you have configured the NVIDIA DRA Driver pack, you can add it to an existing cluster profile, as an add-on profile, or as a new add-on layer to a deployed cluster.
+
+
+## References
 
 - [NVIDIA DRA Driver Documentation](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/dra-intro-install.html)
 - [Kubernetes DRA Documentation](https://kubernetes.io/docs/concepts/scheduling-eviction/dynamic-resource-allocation/)
-- [GitHub Repository](https://github.com/NVIDIA/k8s-dra-driver-gpu)
+- [NVIDIA DRA Driver on GitHub](https://github.com/NVIDIA/k8s-dra-driver-gpu)
+- [NVIDIA GPU Operator Documentation](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/index.html)