All Products
Search

This Product

    Intelligent Computing LINGJUN:Use CNP Performance Evaluation

    Document Center

    Intelligent Computing LINGJUN:Use CNP Performance Evaluation

    Last Updated:Jun 22, 2026

    CNP (Cloud Native Application Performance Optimizer) is a platform for evaluating, analyzing, and optimizing cloud-native application performance. It automates Lingjun cluster training performance evaluation and provides optimization recommendations.

    CNP Platform Entry

    1. Log on to the Lingjun console.

    2. In the left navigation bar, click Performance Evaluation > CNP Performance Evaluation Platform.

    3. In the CNP platform, you can start a performance evaluation or view evaluation results.

    4. In the lower-left corner of the page, click Back to quickly return to the Lingjun console.

    image.png

    image.png

    Start Evaluation

    Step 1: Select a Cluster

    On the welcome page or the Performance Evaluation page, click Start Evaluation to enter the first step: Select a cluster.

    image.png

    • Cluster name: Select the cluster for the evaluation from your available clusters.

    • Authorize DLC access info : After completing the fields, click Test Connectivity. If the connection succeeds, a success message is returned; otherwise, the failure reason is displayed. Common failure reasons:

      Failure reason enumeration

      Suggested operation

      Connection timeout

      Add the current IP to the CNP access whitelist and try again

      Incorrect info

      At least one of AccessID, AccessKey, workspace, or Endpoint is incorrect. Check the info and try again.

      Failed to obtain STS token (D3001)

      Failed to create SLR (D3002)

      Failed to create Arms instance (D3003)

      Failed to check Arms service (D3004)

      Activate ARMS Service

      Failed to retrieve ARMS info (D3005)

      No permission to create SLR (D3006)

      Authorize SLR

    After the connectivity test passes, click Next to proceed to Step 2: Select a test solution.

    Step 2: Select a test solution

    Use Template

    The system provides two default test solution templates. Select one based on your business scenario.

    image.png

    Solution

    Test content

    Scale of the cluster to be tested

    Solution A: General solution for LLM-based scenarios

    • Single GPU test: MatMul (matrix operator)

    • Single machine test: Bert-base

    • AI model test: LLaMA-7B

    • Single GPU test: By default, uses the maximum scale of your cluster.

    • Single machine test: By default, uses the maximum scale of your cluster.

    • AI model test: By default, evaluation jobs are created for 8 GPUs, 16 GPUs, 32 GPUs, 64 GPUs, 128 GPUs, 256 GPUs, and 512 GPUs, limited by the maximum scale of your cluster. For example, if the maximum scale of your cluster is 100 GPUs, evaluation jobs are created only for 8 GPUs, 16 GPUs, 32 GPUs, and 64 GPUs.

    Solution B: General solution for image recognition scenarios

    • Single GPU test: MatMul (matrix operator)

    • Single machine test: Bert-base

    • AI model test: Swin-Transformer and Stable Diffusion

    • Single GPU test: By default, uses the maximum scale of your cluster.

    • Single machine test: By default, uses the maximum scale of your cluster.

    • AI model test: By default, evaluation jobs are created for 8 GPUs, 16 GPUs, 32 GPUs, and 64 GPUs, limited by the maximum scale of your cluster. For example, if the maximum scale of your cluster is 16 GPUs, evaluation jobs are created only for 8 GPUs and 16 GPUs.

    Custom solution

    If the provided templates do not meet your requirements, create a custom test solution.

    1. Single GPU test: The number of nodes can be customized. The default test case is MatMul.

    2. Single machine test: The number of nodes can be customized. The default test case is Bert-base.

    3. AI model test: The model and the number of GPUs in the evaluation cluster can be customized.

    Note

    • Currently supported models include: LLaMA-7B, Stable-Diffusion, Swin-Transformer, Bert-base, and UNet.

    • Default parameter settings use the baseline configuration. View specific configurations on the page.

    image.png

    Estimated Evaluation Time

    After you select a test solution, the estimated evaluation time is calculated based on the test content in the solution. Note: This estimate assumes the maximum scale of the cluster you selected in step one. If your active cluster has not reached its maximum scale, the actual evaluation time will be longer.

    One-Click Start Evaluation

    After completing steps one and two, click One-Click Start Evaluation to start the evaluation.

    View Evaluation Progress and Results

    After a test plan is created, you can view its execution status and progress in real time on the evaluation plan list page. Find the test plan and click Details in the Actions column to view the evaluation progress of each step.

    image.png

    Single GPU Test

    • Test passed

      If no suspected faulty GPUs or warning GPUs are detected among the tested GPUs, the single GPU test is passed.

      Note

      • Suspected faulty GPU: The job that tests the GPU failed, and the GPU may be faulty.

      • Warning GPU: The TFLOPS of the GPU exceeds the normal threshold range for more than 5% of iterations.

      • Logic for calculating the normal threshold range: Take the median TFLOPS of all GPUs in each iteration as the baseline, and compare 103% and 97% of the baseline with 4×sigma (4× standard deviation). The greater values are used as the upper and lower bounds of the normal threshold range.

      image.png

    • Test Results Abnormal

      If at least one suspected faulty GPU or warning GPU is detected among the tested GPUs, the single GPU test returns abnormal results.

      In the evaluation job list, click the plus (+) icon to view details of suspected faulty or warning GPUs. Report abnormal nodes to the O&M team for troubleshooting. Click Evaluation Details to view the evaluation results of this job.

      image.png

    Single machine test progress

    • Test passed

      If no suspected faulty nodes or warning nodes are detected among the tested edge zones, the single machine test is passed.

      Note

      • Suspected faulty node: The DLC job on the node failed, and the node may be faulty.

      • Warning node: The throughput of the node exceeds the normal threshold range for more than 5% of iterations.

      • Logic for calculating the normal threshold range: Take the median throughput of all nodes in each iteration as the baseline, and compare 103% and 97% of the baseline with 4× sigma (4× standard deviation). The greater values are used as the maximum and minimum thresholds of the normal threshold range.

      image.png

    • Abnormal test results

      If at least one suspected faulty edge zone or warning edge zone is detected among the tested edge zones, the single machine test returns abnormal results.

      In the evaluation job list, click the plus (+) icon to view details of suspected faulty or warning nodes, and report abnormal nodes to the O&M team for troubleshooting. Click Evaluation Details to view the evaluation results of the job.

      image.png

    AI model test

    • Test Progress

      Pending: All jobs are in the pending state.

      Completed: All jobs have succeeded, failed, or stopped.

      Stopped: All jobs are stopped.

      Running: Some jobs are complete, and some jobs are pending or running.

      image.png

    • Test job list

      View all jobs included in the AI model step of the current Test Plan. To stop a running job, click Stop. All jobs can be deleted.

      Warning

      Deleted jobs and failed jobs are not included in the performance dashboard. Proceed with caution.

    View test result performance dashboard

    Operation Entry

    You can view the performance dashboard for test plans with a status of Completed. The dashboard covers evaluation jobs that ran successfully in the AI model test phase of the current test plan.

    image.png

    Dashboard Content

    Scalability of Test Model

    image.png

    Shows how throughput changes as the GPU count scales for each model evaluated in the current test plan, reflecting model scalability on the cluster. Results are not compared across different models.

    Formula: Scalability Score = log₂(model throughput / minimum evaluation specification throughput)

    Note

    Example: This example uses the GPT3-175B model (MOCK data, for illustration purposes only).

    GPUs

    Throughput

    Scalability Score

    Theoretical Scalability Score

    64

    10

    128

    18

    log₂(18 / 10)

    log₂ 2

    256

    35

    log₂(35 / 10)

    log₂ 4

    512

    69

    log₂(69 / 10)

    log₂ 8

    1024

    137

    log₂(137 / 10)

    log₂ 16

    Note: The closer the Scalability Score is to the theoretical Scalability Score value, the better the performance scalability.

    Evaluation Result Details

    Shows throughput, MFU, and iteration latency for each model at the GPU counts evaluated in the current test plan. The Y-axis represents the GPU count, and the X-axis represents the metric value.

    image.png