If you are experiencing performance issue, you may be asked by an NVIDIA support agent for an event trace log file. Follow the steps below to generate an event trace log file that our software team can view to better understand the cause of this issue.
- 1. Capturing a GPUView trace requires that the Microsoft "Windows Performance Toolkit" be installed prior. Install the "Windows Performance Toolkit" by downloading the Windows 10 SDK installer from the URL below: https://developer.microsoft.com/en-us/windows/downloads/windows-10-sdk/
- 2. Open the Windows 10 SDK setup file.
- 3. Select "Next" from the "Specify Location" window as shown below.
- 4. Choose from one of the option in the "Windows Kits Privacy" window and then choose "Next".
- 5. Review the "License Agreement" and then choose "Accept" to proceed.
- 6. Uncheck all of the features except for "Windows Performance Toolkit" as shown below. Then choose "Install" to begin the installation process.
- 7. Once completed, click on the "Close" button to complete the installation.
Steps to capture a GPUView trace
1. Open a Command Prompt window with Administrative Privileges by left-clicking over the Windows Start icon on the taskbar -> type "command prompt" to display the launch icon and right-click over the icon as shown below -> select "Run as administrator"
2. From the Command Prompt, navigate to the GPUView directory by typing the following command and then press ENTER:
cd: C:\Program Files (x86)\Windows Kits\10\Windows Performance Toolkit\gpuview
If you chose a different installation path, navigate to that directory and drill down to the gpuview folder.
3. To start logging, type the following command (the first time you run the command, you may need to repeat the process one time to start recording).
4. Keep the Command Prompt window and start the application. This tool can be started at anytime so if your issue occurs at a predictable rate (eg. every 60 seconds, every 10 minutes), wait until you are close to the period when the issue will occur and then start logging. You should not record more than 1 minute of activity because of the size of the log file it will save. If you are capturing a difficult to reproduce stutter, after you record a stutter, please wait at least 10 more seconds before you exit the app to stop recording. Once you have reproduced the issue, go back to the desktop and type the command "log.cmd" again to stop recording. The Command Prompt window should look similar to the screen below:
5. If you find the the final GPUView trace file size to large to upload to a file sharing site when zipped, you may use the command "log.cmd light" instead to start recording a log (you still use just the plain "log.cmd" to stop the capture). We won't get quite as much information in this trace, but more often than not it's still enough to give us a lead.
6. Open Windows Explorer and navigate to the gpuview folder. Several new files will be saved to this folder with the extension .etl. The only file we require is "Merged.etl".
7. Since the Merged.etl is quite big, we recommend compressing the file before you send it to NVIDIA. Windows 10 includes a built in zip utility. To use the built in Windows zip utility, right-click over the Merged.etl file -> from the pop up context menu,select Send to -> select Compressed (zipped) folder
Alternatively you can use the free 3rd party program 7-zip instead. You may download this free program from the URL below:
8. Once you have you have compressed the Merged.etl file and have the new zip file ready, you will next need to upload the zipped file to a file sharing site such as Google Drive or Microsoft OneDrive.
9. If you are working with an NVIDIA Customer Care support agent, you may provide the agent with the URL to download the GPUView ETL trace file. If you are providing driver feedback through the NVIDIA Display Driver Feedback form, you may email the URL to download the GPUView ETL trace file to email@example.com. In your email, please include the name of the app the log was taken from and basic PC specs (CPU make and model/GPU make and model/system RAM/number of monitors/operating system).