> ## Documentation Index
> Fetch the complete documentation index at: https://docs.scitix.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# View Resource Usage

> View resource pool usage, user GPU usage, idle task duration, node disks, and GPU node failures in the resource dashboard to determine whether resources are constrained.

When you need to determine whether resources are sufficient in a specific region, cluster, instance specification, or resource pool, you can view resource usage in the Resource Dashboard. The Resource Dashboard provides a resource-centric view of quotas, usage, scheduling rates, unavailable resources, user usage, and GPU node failures, making it suitable for capacity assessment, resource allocation, usage reviews, and failure impact assessment.

## Select Tabs by Analysis Target

After navigating to **Resources > Resource Dashboard**, select a tab based on your analysis objective.

| Tab | Purpose |
| - | - |
| **Resource Pool** | View changes in provisioned, occupied, and unavailable resources. |
| **User GPU Usage Ranking** | Compare users' GPU card-hours, GPU utilization, and task composition, and access user resource details to analyze idle-time task duration. |
| **User Usage History** | View user usage trends in the resource pool. |
| **Disk** | View node disk capacity, remaining writable space, top-consuming Pods, and 24-hour eviction status. |
| **GPU Node Fault** | View historical repair durations, failure-duration levels, source contributions, and real-time failed nodes for GPU node failures. |

When comparing different tabs or exported data, ensure that the filter criteria—such as region, cluster, resource type, resource pool, and time range—are consistent. If the page provides a project group filter, use the same project group scope as well to avoid values that cannot be directly compared due to differing statistical scopes.

## View Resource Pool Usage

Go to **Resources > Resource Dashboard**, open the **Resource Pool** tab, and select a region, cluster, resource type, instance specification, and time range to view the information.

The resource pool tab is used to view trends and summary information by resource.

![Resource Pool Usage](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/resource-pool-usage-dashboard.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=7770f7cfd9a7e4230d1c2eaa2cf95977)

| Information | Purpose |
| - | - |
| Total quota and usage | Determine the resource capacity and utilization of the current resource pool or instance specification. |
| Scheduling rate | Determine whether resources are being scheduled effectively. |
| Unavailable resources | Determine whether any resources cannot be used by tasks. |
| Trends over a time range | Observe whether resource utilization continues to increase or decrease during a specific period. |
| Specification filter | Focus on resource usage for a specific GPU / CPU instance specification. |

If the usage of a resource type remains close to its total quota, or a large number of resources are unavailable, subsequent tasks may be queued or fail to be submitted. You can continue to view the resource pool details to determine whether you need to scale out, adjust the resource pool configuration, or renew resource orders.

## View User GPU Usage Ranking

Go to **Resources > Resource Dashboard** and open the **User GPU Usage Ranking** tab to view resource usage within a specified time range by user. You can filter by user, region, cluster, specification, resource type, resource pool, task type, task scale, and time range.

![User GPU Usage Ranking](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/user-resource-usage-ranking.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=a1e804da3b4204dda23ad0fec1c1b290)

| Information | Purpose |
| - | - |
| **User** | Identifies the resource consumer. |
| **GPU SM Utilization** | Shows the user's GPU compute utilization. |
| **GPU Hours Used** | Measures GPU hours consumed by the user during the selected period. |
| **Task Type Distribution** | Shows the workload types that account for most of the user's resource consumption. |
| **Scale Distribution** | Shows the scale of the user's workloads. |
| **GPU Type Distribution** | Shows which GPU types the user uses most frequently. |
| **Submitted Tasks** | Shows the number of tasks submitted by the user during the selected period. |
| **Quota** | Shows the user's quota. |
| **Quota Used** | Shows how much of the user's quota has been used. |

Statistics in **User GPU Usage Ranking** are cached for 30 minutes. The page displays the results of the previous calculation. Tasks that have just been submitted or completed may not be included in the statistics immediately.

To analyze the resource consumption of an individual user, click the target user in **User GPU Usage Ranking** to go to the **User Resource Detail** page. When comparing different charts, keep the user, region, cluster, specification, type, resource pool, and time range consistent.

![User Resource Detail](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/user-resource.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=d77acd7d97d8b71218b2acbd711bfbc0)

| Information | Purpose |
| - | - |
| **Allocated GPU Hours** | Displays the GPU hours consumed by the user within the selected statistical range. |
| **SM Utilization** | Shows the user's GPU SM utilization. |
| **Submitted Tasks** | Shows the number of tasks submitted by the user. |
| **Resource usage (GPUs)** | Displays changes in the user's resource usage over time. |
| **SM utilization (%)** | Compares SM utilization for guaranteed and idle resources. |
| **Workload scale (GPU hours)** | Displays the distribution of resource consumption by workload scale. You can switch between **GPU hours** and **Workloads**. |
| **Workload type (workloads)** | Shows the number of workloads by type. |
| **Idle task queue duration (minutes)** | Shows the queue duration of idle tasks. |
| **Idle task run duration (minutes)** | Shows the run duration of idle tasks. |
| **Resource pool / GPU model × workload type** | Breaks down usage by resource pool or GPU model and workload type. The table can be switched between **Resource pool** and **GPU model**. |

Data in **User Resource Detail** is also cached for 30 minutes. Tasks that have just been submitted or completed may not be immediately reflected in the aggregated metrics and charts.

## View User Usage History

Go to **Resources > Resource Dashboard**, open the **User Usage History** tab, select one or more users, then select a region, cluster, resource type, resource pool, and time range to view changes in user usage within the specified resource pool.

![User Usage History](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/user-usage-history.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=5e74cb1337f193370936145c18589fee)

User usage history displays weighted average usage over the selected time range and is suitable for observing long-term trends. The page provides the following information:

| Information | Purpose |
| - | - |
| User filter | Select one or more users to compare changes in resource usage across users. |
| Resource pool filter | Focus on a specified resource pool to determine whether resources are continuously occupied by a small number of users. |
| **Weighted Avg Usage** | View the selected users' weighted average usage within the selected resource pool. |
| Time range trend | Observe usage changes during capacity expansions, quota adjustments, or team review cycles. |

If you need to confirm real-time changes immediately after a task releases resources, prioritize viewing the task details or resource pool details. If the page prompts you to select a user and resource pool, complete the filtering before viewing historical trends.

## View Disk Usage

Go to **Resources > Resource Dashboard**, open the **Disk** tab, and filter by region, cluster, resource pool, and node to view node disk usage.

![Disk usage in Resource Dashboard](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/resource-dashboard-disk.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=b5ae7c526a86c563efe11dc59a321fe6)

The **Disk** tab lists nodes in ascending order of **Writable Remaining** space, making it easier to prioritize nodes with less remaining writable space. The page provides the **Warning Only (>80%)** and **Evicted in 24h** filters to focus on nodes with high utilization or evictions within the last 24 hours.

| Information | Purpose |
| - | - |
| **Node** | Identify the node corresponding to the disk usage. |
| **Resource Pool** | Identify the resource pool that contains the node. |
| **Cluster** | Identify the cluster that contains the node. |
| **RoCE Pod** | Identify the RoCE Pod associated with the node. |
| **Status** | Determine whether the node disks are in a normal state. |
| **Used** | Assess the disk space currently in use. |
| **Capacity** | Check the node disk capacity. |
| **Writable Remaining** | Assess how much writable disk space remains. |
| **Top Consuming Pods** | View the Pods with the highest disk usage; hover to view the top five Pods and their associated users. |
| **24h Evictions** | Check whether any evictions occurred within the past 24 hours. |

If the node has low remaining writable space, or the **24h Evictions** value is not 0, use **Top Consuming Pods**, task details, and task resource usage to further identify the specific workload.

## View GPU Node Faults

Go to **Resources > Resource Dashboard**, open the **GPU Node Fault** tab, and filter by instance specification, region, cluster, and time range to view historical trends and real-time status for GPU node failures.

| Area | Purpose |
| - | - |
| **Historical Trends and Statistics** | Based on the currently selected time window, region, cluster, and resource pool scope, view **Completed Repair Samples**, **Average Repair Duration**, **P90**, and **P95**. |
| **Historical Repair Data** | View time trends for the average, P90, and P95. Historical data is for reference only and does not constitute a repair time commitment. Actual repair duration may be affected by failure type, spare-part availability, and vendor response time. |
| **Failure-Duration Level Trend** | View the failure-duration level for entire machines at 12-hour intervals to determine the proportion of failed machine duration relative to total customer machine duration. Calculation: level = failed machine duration / total customer machine duration. |
| **Failure Duration Contribution** | Break down failure duration by `cordon` source to distinguish between **Automatic Detection** and **Manual Reporting**. Calculation: contribution rate = failure duration for this category / total customer duration. |
| **Real-Time Data Dashboard** | View the latest status of all customer resources, including **Failure Duration Level - Entire Machine Level**, **Number of Failed Machines**, and **Alerted · Pending Clearance Approval**. Real-time data is not affected by the historical filter conditions above. To view the latest status, click **Refresh**. |
| **Failed Node Details** | View the region, cluster, resource pool, node name, instance type, GPU model, failure cause, and failure duration of failed nodes. If there are currently no failed nodes, the table displays an empty state. |

## Determining Whether Resources Are Constrained

When a task cannot start or remains queued for an extended period, the Resource Dashboard can help you first determine whether capacity is constrained or resources are heavily concentrated in use:

| Diagnostic Goal | How to Check |
| - | - |
| Whether the resource pool is full | On the **Resource Pool** tab, view the total quota and used capacity for the target instance specification. |
| Whether unavailable resources exist | On the **Resource Pool** tab, view unavailable resources to determine whether any resources cannot be used by tasks. |
| Whether a small number of users have occupied resources for an extended period | In **User GPU Usage Ranking** and **User Usage History**, view the primary sources of resource consumption and their duration. |
| Whether node disk capacity is constrained | On the **Disk** tab, view **Writable Remaining**, **Top Consuming Pods**, and **24h Evictions**. |
| Whether GPU node failures exist | On the **GPU Node Fault** tab, view historical repair durations, failure-duration levels, and real-time failed nodes. |

For a complete investigation into why a task cannot start or remains queued for a long time, see [Troubleshoot insufficient task resources or long queue times](/siverse/resources/troubleshoot-insufficient-resources-or-queued-tasks.md).

## Export Resource Data

For capacity analysis or team reconciliation, you can export resource usage data from the Resource Dashboard based on filter criteria.

Before exporting, we recommend confirming that the region, cluster, resource type, instance specification, and time range align with the analysis criteria. Retain the filter scope, statistical units, and time window in the analysis records. To explain resource peaks, you can also review task resource utilization. If you only need to verify a recently completed create, stop, or scale operation, prioritize checking the task details, resource pool details, or order status.

## Frequently Asked Questions

> **Why do the figures in the Resource Dashboard and resource details differ?**
>
> First, verify that the region, cluster, resource pool, instance specification, unit of measurement, and time range are consistent across both pages. Some ranking and trend data may be subject to statistical caching, so recently submitted, stopped, or released tasks may not be reflected in the dashboard immediately. If the page is blank, displays an error, or GPU SM utilization is 0, also confirm that monitoring samples are available. If discrepancies remain after aligning the criteria, retain screenshots of both pages, the query time, and field names, and contact platform support.

## Related Documentation

* [View task resource usage](/siverse/resources/view-task-resource-usage.md)
* [Troubleshoot insufficient task resources or long queue times](/siverse/resources/troubleshoot-insufficient-resources-or-queued-tasks.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.