Running on Linux

Purpose

This section serves to:

Overview

In order to analyze Vulkan applications on Linux, one must leverage GPU PerfStudio's networked client/server model.

For the user, this implies two things:

This requirement exists because the GPU PerfStudio client only supports Windows.

Usage Example

1) Linux machine: Make note of GPUPerfServer location. In this example, it is /home/test/Desktop/GPUPerfStudio/GPUPerfServer

vk_linux_1.png

2) Linux machine: Open a terminal and go to the application's working directory. In this example, it is /home/test/Desktop/apps and the application is tri

vk_linux_2.png

3) Linux machine: From the application's working directory, launch it with GPUPerfServer using the --vk flag. In this example, the full command would be: /home/test/Desktop/GPUPerfStudio/GPUPerfServer --vk ./tri

vk_linux_3.png

4) Linux machine: At this point the app should be running normally, alongside an active GPUPerfServer terminal window.

vk_linux_4.png

5) Windows machine: Launch GPUPerfClient.exe, and press the blue Connect button.

6) Windows machine: Set the Session Type to Remote, specify server IP, and specify port (default 8080)

vk_remote_conn.png

7) Windows machine: Press Next, followed by Connect, and an API selection dialog should appear with the Vulkan application listed. Select it and press OK.

vk_api_dialog.png

8) Windows machine: At this point, the client-server connection is established and the application may now be analyzed.

9) Windows machine: Press either the API Trace, GPU Trace, or Linked Trace button to begin analysis.

Best Practices

It is recommended that users first learn GPU PerfStudio using a simple Vulkan application. Once confident that the basics work, one should then proceed to analyze more complex applications.

Run the application on its own at least once, prior to running it through GPU PerfStudio.

Repeated experimentation is encouraged.

Users are not limited to one set of data per session. Users may refresh current data by repeatedly pressing the API Trace, GPU Trace, or Linked Trace buttons on the toolbar.

The default behavior is to show information for a single frame, but GPU PerfStudio has the ability to collect data across N frames. This number can be specified in the Settings pane, under Number of Frames to Capture.

Analysis of Steam Applications

Sometimes applications may not be launched directly from their executable. This is true for some applications which are launched from Steam. In this case, users must use GPUPerfServer to launch steam.sh itself, instead of the application executable. After Steam loads, use it to launch the application as normal. The application should then be visible to GPU PerfStudio, and will be listed in the API selection dialog. At that point it will be ready for analysis.

Known Limitations