Introduction
This site covers the NVIDIA GPU vendor base path for ACP accelerator workloads. It helps users prepare NVIDIA GPU nodes, install and operate the GPU host software stack, and make GPUs available to Kubernetes workloads.
Alauda GPU Management is the core product in this base path. It installs and manages the NVIDIA GPU software stack on ACP workload clusters from a single ClusterPolicy custom resource: the GPU driver, the container runtime integration, the NVIDIA device plugin, GPU Feature Discovery, and DCGM metrics.
Use this product when ACP workloads need to request NVIDIA GPUs as the nvidia.com/gpu extended resource on the nodes where the device plugin reports them.
Unified delivery
Earlier releases delivered these capabilities as separate cluster plugins that each had to be installed and upgraded on their own — a standalone NVIDIA GPU Device Plugin, a standalone DCGM-Exporter, and out-of-band driver and container runtime setup. Alauda GPU Management replaces that model: one OLM operator and one ClusterPolicy reconcile the driver, container-toolkit, device plugin, GPU Feature Discovery, and DCGM-Exporter together.
Because the product now tracks the upstream GPU Operator line rather than the device-plugin line, the version advances to v26.3.x. For the mapping from the previous standalone components to the unified operator, see Versions and Components.
When to use it
Use Alauda GPU Management when:
- you run container workloads on NVIDIA GPU nodes and want them scheduled through
nvidia.com/gpu; - you want the GPU driver, container runtime integration, device plugin, feature discovery, and DCGM metrics managed as one ACP product instead of separately installed plugins;
- your GPU nodes run an immutable operating system (such as SUSE Linux Micro) and need a precompiled driver that keeps the host image unchanged;
- your GPU nodes already have a host driver and you only need the operator to manage the Kubernetes-facing GPU stack.
Use the HAMi documentation when the task starts from GPU sharing, memory or compute partitioning, or HAMi scheduling behavior on NVIDIA GPUs. In that combined path, this site provides the NVIDIA GPU base — driver, runtime, and node readiness — while HAMi documentation owns HAMi installation, device exposure, sharing semantics, and its own resource keys.
Use the ACP accelerator pages when you only need to choose between the GPU, NPU, HAMi, DRA, and VM GPU passthrough paths. Use NVIDIA's own documentation for vendor hardware concepts, driver internals, and CUDA details that do not change ACP operation.