> ## 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 Task Resource Usage > View resource usage for running and historical tasks in Task Dashboard, and review inference service metrics. The Task Dashboard provides a workload-based view of resource pools, resource usage, and GPU/CPU utilization. Use it to identify which tasks or users are consuming resources and to investigate low-utilization tasks. ## Select a Tab by Analysis Target After navigating to **Resources > Task Dashboard**, select a tab based on your analysis objective. | Tab | Purpose | | - | - | | **Running tasks** | View currently running workloads and their resource usage and utilization. You can open task details or stop and delete tasks. | | **All tasks** | Query tasks submitted within the past 14 days and their history. Find tasks by status, user, resource pool, or other criteria. | | **Idle Task Monitoring** | Review idle training task submissions, GPU hours, run duration, preemption, and participating users. | | **Inference Service Monitor** | View trends in latency, requests, throughput, cache hit rate, and token-length distribution for an inference service. | ## View Running Tasks Open the **Running tasks** tab. Use the region and cluster selectors, the **All Types**, **All Specs**, and **All Users** filters, the resource pool filter, and the task-name search field to find tasks. ![Running tasks in the Task Dashboard](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/task-resource-usage-dashboard.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=b90b63a330717bbbf94ccfede7c4d175) ### View Resource Usage The task list provides the following resource and ownership information: | Information | Purpose | | - | - | | **Name**, **Type**, **User** | Identify the task and confirm its type and submitting user. | | **Region**, **Cluster**, **Resource Pool** | Confirm the resource scope in which the task resides. | | **Resource Type**, **Instance Spec**, **GPU Model**, **Usage** | View the resource configuration requested and occupied by the task. | | **GPU Resource Utilization** and **CPU Resource Utilization** | Determine whether the task is fully using its requested resources. | > [!TIP] > > * Use **List Sort** and select **Submit Time: Newest First** to sort by submission time. > * Under **Utilization Metric**, select **avg**, **min**, or **max**, then choose a time range. > * Click **Columns** to adjust the fields displayed in the list. When a task cannot start or remains queued for an extended period, check whether the target resources are occupied by other tasks: | Objective | How to Check | | - | - | | Whether the target resource pool is already occupied by tasks | Filter running tasks by resource pool. | | Which users are occupying resources | Review the users and resource usage in the task list. | | Whether there are low-utilization tasks | Compare task resource usage with GPU/CPU utilization. Open task details for further investigation. | For further troubleshooting of tasks that fail to start or remain queued for an extended period, see [Troubleshoot insufficient task resources or long queue times](/siverse/resources/troubleshoot-insufficient-resources-or-queued-tasks.md). ### Handling Low-Utilization Tasks If a task is consuming resources but GPU/CPU utilization remains low for an extended period, first check the following: * Whether the task is still in an active running phase. * Whether data reading, the network, or storage has become a bottleneck. * Whether the task specifications significantly exceed actual requirements. * Whether the training task has communication waits, node exceptions, or uneven load distribution. * Whether the number of inference service replicas exceeds actual traffic requirements. After confirming that a task is no longer needed, you can stop it in the task list or the relevant business module. Stopping a task releases the instance resources it occupies, but may interrupt training or online services. ### Manage Running Tasks | Actions | Description | | - | - | | **Detail** | Open the task details page to view real-time utilization and task runtime information. | | **Stop** | Stop tasks that no longer need to run. Supports single and batch operations. | | **Delete** | Delete the target tasks. Supports single and batch operations. | > [!NOTE] > > Stopping or deleting affects task execution. Before proceeding, verify the target task to avoid accidental actions. ## View All Tasks Open the **All tasks** tab to view tasks submitted in the past 14 days and their history. Filter by region, cluster, workload type, resource type, resource pool, instance specification, user, and status, or search by name. ![All tasks in Task Dashboard](https://mintcdn.com/scitix-adfc65c2/ZqwdmfntBUDyGPkJ/siverse/resources/media/task-all-tasks.png?fit=max\&auto=format\&n=ZqwdmfntBUDyGPkJ\&q=85\&s=7efa7f2ee073f43eb6e275b9e56b7528) Workload type is single-select. Task status supports multiple selections and displays the corresponding original statuses for the selected workload type. The list provides the following information: | Information | Purpose | | - | - | | **Name**, **Type**, **User** | Identify the task and confirm its type and submitting user. | | **Resource Pool**, **Resource Type** | Confirm the resource scope and resource type used by the task. | | **Instance Spec**, **GPU Model**, **Usage** | View the instance specification, GPU model, and quantity used by the task. | | **Created At (UTC)**, **End time (UTC)**, **Status** | View when the task ran and its result. | ## View Idle Task Monitoring Open the **Idle Task Monitoring** tab to review resource usage, run duration, and preemption for idle training tasks. This tab reports monitoring data for idle training tasks only. ### Review Key Metrics First select a region, cluster, and time range. The summary at the top of the page reports key metrics for the selected filters: | Metric | Purpose | | - | - | | Submitted tasks | Number of idle training tasks submitted during the selected period. | | GPU hours | GPU hours consumed by idle training tasks. | | Average run duration | Average run duration of idle training tasks. | | Time to first preemption (P50) | Typical time an idle training task runs before its first preemption. | | Preempted tasks / total idle tasks | Number of preempted tasks and the total number of idle tasks in the selected period. | | Preemption rate | Proportion of idle training tasks that were preempted. | ### Review Idle Task Trends Idle task trends show how idle resources are used during the selected period. The page typically includes these charts: | Trend | What to Look For | | - | - | | Idle task submissions | Changes in idle training task submissions over time, including submission spikes or recurring fluctuations. | | Mean / P50 run duration before first preemption | How long idle tasks typically run before their first preemption, which can indicate how stable idle resources are. | Use these trends together with the summary metrics: * If submissions increase but GPU hours do not rise at the same pace, tasks may be running for shorter periods or being preempted sooner. * If the preemption rate increases while the time to first preemption decreases, idle resources may be available for shorter periods. * If the mean and P50 run durations differ significantly, task run durations may vary widely. Review individual tasks for more detail. Trend data describes overall idle task behavior and does not show the real-time status of an individual task. To confirm whether a task was preempted, open its details or check its status in the training task list. ### Review Participating Users The participating users list shows which users submitted idle training tasks during the selected period. It typically includes: | Information | Purpose | | - | - | | User | Identify the user who submitted idle training tasks. | | Submitted tasks | Number of idle training tasks submitted by the user during the selected period. | | Pod count | Number of Pods used by the user's idle training tasks. | | Idle resources requested | Scale of idle resources requested by the user. | If a user's submission count or requested idle resource amount is notably high, review the corresponding tasks under **Running tasks** or **All tasks**. Then determine whether to adjust the submission rate, task specifications, or checkpoint strategy. ## Review Inference Service Metrics Open the **Inference Service Monitor** tab, select the region, cluster, inference service, and time range to view trends for the following metrics: | Metric | Purpose | | - | - | | Time to First Token (TTFT) (s) | Monitor the latency for a request to receive its first token. | | Time per Output Token (TPOT) (s) | Monitor the time between tokens during generation. | | Running requests | View changes in the number of requests being processed by the service. | | Token throughput (TPM) (tok/min) | View changes in the number of tokens processed per minute. | | Request rate (RPM) (req/min) | View changes in the number of requests processed per minute. | | Prefix cache hit rate (%) | Monitor the proportion of prefix cache hits. | | Input token length distribution | View the distribution of input token lengths. | | Output token length distribution | View the distribution of output token lengths. | ## Related Documentation * [View resource usage](/siverse/resources/view-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.