GPU Card Management
Overview
The GPU Card Management page provides a unified view of all physical GPU cards in the cluster, supporting filtering and viewing GPU resource allocation and usage by node, vendor, model, and status.
Viewing the GPU Card List
In the left navigation pane, select Node Management → GPU Cards to access the cluster GPU card list.
Overview Panel
The top of the page displays the cluster GPU resource status:
- Total GPU Count: Total number of physical GPU cards in the cluster
- Health Status Distribution: Pie chart showing the ratio of healthy vs. unhealthy GPU cards
- Vendor/Model Distribution: GPU card counts by vendor and model
GPU Card Table
The table displays detailed information for each GPU card:
| Column | Description |
|---|---|
| Node | Name of the compute node where the GPU is installed; click to navigate to node details |
| Card Index | GPU index number within the node |
| Card UUID | Globally unique GPU identifier; click to navigate to GPU card details |
| Status | Healthy or Sub-healthy |
| Model | GPU model, e.g., A100, V100, 910B |
| NUMA | NUMA node number the GPU belongs to |
| Vendor | GPU vendor: NVIDIA, Huawei Ascend, Cambricon, etc. |
| Virtualization Mode | Current GPU virtualization running mode |
| vGPU | vGPU used / total |
| Compute | GPU compute used / total |
| Memory | VRAM used / total (GiB) |
Filtering and Search
The following filters are supported:
- Filter by Node: Select a specific node from the dropdown
- Search by UUID: Enter a GPU UUID for exact matching
- Label Selector: Filter using Kubernetes labels
- Filter by Vendor: Select NVIDIA, Huawei Ascend, Cambricon, etc.
- Filter by Model: Enter GPU model keywords
- Filter by Status: Healthy / Sub-healthy
Viewing GPU Card Details
Click a GPU card's UUID in the list to navigate to its detail page.
Basic Attributes
| Attribute | Description |
|---|---|
| Status | Current health status |
| Node | Physical node where the GPU is installed |
| Card UUID | Globally unique identifier |
| Card Index | Index within the node |
| Model | GPU model |
| Vendor | GPU vendor |
| NUMA | NUMA node number |
| Virtualization Mode | Virtualization mode inherited from the node |
| Total Memory | Physical VRAM size (GiB) |
| Driver Version | GPU driver version (if available) |
| Max Device Split Count | Maximum concurrent tasks after vGPU slicing (if configured) |
Resource Summary
Displays current GPU card resource usage:
- Compute: Used / Total
- Memory: Used / Total (GiB)
- vGPU: Allocated / Total
Detail Tabs
| Tab | Content |
|---|---|
| Running Status | Real-time monitoring charts for GPU utilization, memory usage, allocation rate, power, and temperature, plus GPU topology |
| Pods | List of all Pods currently scheduled to this GPU |
| Monitoring | Historical monitoring metric charts with custom time range support |
| Events | Kubernetes events related to this GPU |
GPU Cards in Node Details
In Node Management → Cluster Nodes, click a node to enter its detail page. For nodes equipped with GPUs, the running status page displays:
- Total GPUs: Number of GPU cards on the node
- Total VRAM: Combined VRAM across all GPUs on the node
- Total vGPUs: Total vGPU count on the node
- GPU Card List: Status, model, vGPU/compute/memory usage for each GPU card
Pod Scheduling Information
For Pods using GPU resources, the Pod detail page shows Scheduled to GPU information, including:
- Allocated GPU UUID with a link to the physical GPU card
- Allocated vendor type
- Allocated VRAM size
- Allocated compute percentage
Previous
