Analysis Summary

The Analysis Summary page consists of two parts: Analysis Summary and Collection And Platform Info.
  • On the Analysis Summary page, you need to refer to Merge Reports to import the parent directory of PROF_XXX to display the collected communication duration proportion data of all NPU nodes in the cluster scenario, as shown in Figure 1.
  • On the Collection And Platform Info page, you can select ID of the target device to view its detailed hardware and profiling information, as shown in Figure 2.
Figure 1 Analysis Summary
Figure 2 Collection And Platform Info

Analysis Summary

Table 1 Analysis summary

Field

Description

Analysis Summary

Analysis summary.

Bottlenecks And Profiling Suggestion

Bottlenecks and profiling suggestions.

The communication time ratio of all NPU cards exceeds the threshold of 10%.

The communication duration proportion of every NPU node is greater than 10%.

The communication time ratio of all NPU cards is good and below the threshold of 10%.

The communication duration proportion of every NPU node is lower than 10%.

The communication time ratio of some NPU cards exceeds the threshold of 10%. You need to check whether there are slow nodes or slow links.

The communication duration proportions of some NPU nodes are greater than 10%. Check whether slow nodes or links exist.

Click the Cluster Iteration Analysis tab to obtain more information.

Click Cluster Iteration Analysis for more information.

Top

Top N NPU nodes with the largest communication duration proportions.

Apply

Data export button. If you select a top N value and click this button, the bar chart of top N NPU nodes with the largest communication duration proportions is exported.

Ratio Of The NPU Card

Chart of communication duration proportions of NPU nodes.

Ratio (%)

Communication duration proportions of NPU nodes.

Rank *

ID of each NPU node in the cluster.

Profiling Info

Table 2 Information collection

Field

Description

Result Size

Size of a result file.

Profiling Elapsed Time

Duration of information collection.

Host System Info

Table 3 Host system information

Field

Description

Cpu Num

Number of CPUs.

Host Operating System

Host OS information.

Host Computer Name

Host computer name.

Host CPU Info

Table 4 Host CPU information

Field

Description

CPU ID

CPU ID.

Name

CPU name.

Type

CPU model.

Frequency

CPU frequency.

This parameter is not displayed in some systems because these systems do not have calling frequency APIs. The value of this parameter is subject to the actual situation.

Logical CPU Count

Number of logical CPUs.

Device Info

Table 5 Device information

Field

Description

AI Core Number

Number of AI Cores.

AI CPU Number

Number of AI CPUs.

Control CPU Number

Number of Ctrl CPUs.

Control CPU Type

Ctrl CPU type.

Device Id

ID of the device associated with the current page.

TS CPU Number

Number of TS CPUs.

DDR

Table 6 DDR parameters

Field

Description

Metric

Bandwidth (MB/s)

Read(MB/s)

Read bandwidth (MB/s)

Write(MB/s)

Write bandwidth (MB/s)

AI Core Utilization

The AI Core utilization is displayed in a line chart (displayed only when the sample-based mode is selected).