### Sharing Mode

Volcano-vgpu supports two types of device-sharing: HAMi-core and dynamia-mig, A node can either using HAMi-core, or Dynamic-mig. Heterogeneous is supported(a part of node using HAMi-core, the other using Dynamic-mig)

A brief introduction about these two modes:
HAMi-core is a user-layer resource isolator provided by HAMi community, works on all types of GPU.
Dynamic-mig is a hardware resource isolator, works on Ampere arch or later GPU.
The table below shows the summary:
| Mode        | Isolation        | MIG GPU Required | Annotation | Core/Memory Control | Recommended For            |
| ----------- | ---------------- | ---------------- | ---------- | ------------------- | -------------------------- |
| HAMI-core   | Software (VCUDA) | No               | No         | Yes                 | General workloads          |
| Dynamic MIG | Hardware         | Yes              | Yes        | MIG-controlled      | Performance-sensitive jobs |
You can set the sharing mode and customize your installation by adjusting the configs
Enabling GPU Support in Kubernetes
Once you have enabled this option on all the GPU nodes you wish to use,
you can then enable GPU support in your cluster by deploying the following Daemonset:
Configure and install with Helm
Add the volcano-vgpu-device-plugin Helm repository:

Install the chart:

If you want to enable CDI, you can use the following command to install.

#### Install with yaml

##### Normal Mode

##### CDI Mode
###### Modify deployments/static/volcano-vgpu-device-plugin.yml
Add the following environment variables to the env section of the volcano-device-plugin container.

• name: DEVICE_LIST_STRATEGY

• name: NVIDIA_DRIVER_ROOT

• name: NVIDIA_CDI_HOOK_PATH

• name: GDRCOPY_ENABLED

• name: GDS_ENABLED

• name: MOFED_ENABLED

2. Add the following configuration to the volumeMounts section of the volcano-device-plugin container.

• name: cdi-root

• name: driver-root

• name: usrbin

3. Add the following configuration to the volumes section.

• name: driver-root

• name: cdi-root

• name: usrbin

###### Deploy

### Testing without real GPUs (nvml-mock)

For local development or CI without physical GPUs, the device-plugin can run
against nvml-mock, which provides a
fake libnvidia-ml.so that simulates GPU devices on a kind cluster. The same
approach is used by the HAMi nvml-mock lab.

> Note: nvml-mock only simulates the device discovery path. Use it to verify
> scheduling and registration behavior — the monitor container is automatically
> disabled because it cannot resolve the mock NVML library (SetNvmlLibraryPath
> is only called by the device-plugin binary, and the monitor container does
> not mount the mock driver root), so it would fail NVML initialization.
>
> Note: nvmlMock.enabled and cdi.enabled are mutually exclusive. The
> nvml-mock driver root would override the CDI driver root, so the Helm chart
> fails at render time if both are set to true.

1. Deploy nvml-mock
Deploy nvml-mock first so the fake driver and libnvidia-ml.so are installed on
the node (default host path: /var/lib/nvml-mock/driver):

#### 2. Deploy the device-plugin with nvml-mock (Helm)

nvml-mock support is provided through the Helm chart only, so that
KLOG_LEVEL, DEVICE_CONFIG_NAMESPACE, and other plugin settings stay in sync
with the rest of the deployment:

Adjust the driver root if your nvml-mock install uses a non-default path:

When nvmlMock.enabled=true, the chart injects the NVIDIA_DRIVER_ROOT,
CONTAINER_DRIVER_ROOT, NVIDIA_DEV_ROOT, DEVICE_DISCOVERY_STRATEGY=nvml,
and DP_DISABLE_HEALTHCHECKS=all environment variables, mounts the mock driver
root, and disables the monitor container.

3. Verify
Once the device-plugin pod is running, check the node capacity — a fake
volcano.sh/vgpu-number resource should appear, same as with real GPUs:

The exact number equals (fake GPUs from the nvml-mock profile) ×
deviceConfig.nvidia.deviceSplitCount. The nvml-mock chart's default gb300
profile exposes 4 fake GPUs, so with the default deviceSplitCount: 10 the
node reports 40; a profile with 8 GPUs (e.g. a100) reports 80.

Verify environment is ready
Check the node status, it is ok if volcano.sh/vgpu-number is included in the allocatable resources.

### Running VGPU Jobs

VGPU can be requested by both set "volcano.sh/vgpu-number" , "volcano.sh/vgpu-cores" and "volcano.sh/vgpu-memory" in resource.limit

You can also collect the GPU utilization, GPU memory usage, pods' GPU memory limitations and pods' GPU memory usage metrics on nodes by visiting the following addresses:

Example: