NVIDIA

The NVIDIA vendor integration covers NVIDIA GPU enablement in ACP clusters, including direct GPU allocation, DRA-based GPU allocation, GPU Operator, and NVIDIA GPU metrics components such as DCGM-Exporter when those plugin products are available from the application marketplace or an approved delivery package.

Choose an NVIDIA path

OptionUse when
NVIDIA GPU Device PluginYou need physical NVIDIA GPU resources exposed as Kubernetes extended resources.
NVIDIA DRA Driver for GPUsYou need Kubernetes DRA resource claims for NVIDIA GPU allocation.
GPU OperatorYou need a unified NVIDIA stack deployment path, such as driver, toolkit, device plugin, or monitoring components, when that product is available in your delivery.
DCGM-ExporterYou need NVIDIA GPU metrics for ACP monitoring, billing, or product dashboards.
HAMi on NVIDIA GPUYou need sharing, virtualization, or scheduling behavior built on NVIDIA GPU prerequisites. Use the HAMi sharing and scheduling path for the HAMi part of the workflow.
WARNING

Choose one GPU allocation path for each physical NVIDIA GPU pool.

Do not let multiple GPU allocation components manage the same physical GPUs unless that topology is explicitly documented and validated for your product versions.

Use this page to choose an NVIDIA integration path. Use the corresponding product documentation for installation, node labels, runtime prerequisites, compatibility, and troubleshooting. If HAMi exposes the GPU resources, use this path only for NVIDIA prerequisites and continue with HAMi documentation for resource requests and sharing semantics.

For generic plugin installation, see Cluster Plugin.

Product documentation

NVIDIA GPU documentation

Note
Because Alauda NVIDIA GPU releases on a different cadence from Alauda Container Platform, the Alauda NVIDIA GPU documentation is now available as a separate documentation set at Alauda NVIDIA GPU.

GPU Operator

Use the GPU Operator product documentation when that product is available in your delivery.