Cluster List
On Cluster List screen, the entire GPU infrastructure is displayed as multiple cluster cards. Each card represents an independent Kubernetes cluster, organized by purpose such as development, training, or production.
Clusters are arranged in a grid format. Each cluster is shown as a separate card, with distinct colors and icons that give an intuitive overview of its health status. Clusters with issues are highlighted in red or orange, making them easy to identify and address promptly.
Access GPU Monitoring Screen
From the System Admin Control Panel, click GPU in the main navigation. Then, in the left-side navigation, select GPU Monitoring.
Each cluster is represented as an individual card containing the following information:
| Item | Meaning |
|---|---|
| Available GPUs | Total GPUs and Allocatable GPUs in the Cluster |
| Nodes | Total number of GPU nodes and number of nodes available for GPU allocation |
| Status Mini Map | Display the status of each GPU as a color minimap: • Blue: Available • Yellow: In use • Red: Error |
| Allocation Rate | The percentage of GPUs currently in use out of the total GPUs. |
| Efficiency | The average efficiency based on GPU status. |
| Node saturation | The average total GPU resource usage per node. |
| Memory bandwidth | The average memory utilization rate of the GPUs in use. |
| Power | The total power consumption of all GPUs. |
| ECC | Error counter occurred in GPU memory |
| S (Session) | The number of double-bit errors that have occurred while the GPU is running. |
| L (Lifetime) | The cumulative number of double-bit errors over the entire lifetime of the GPU. |