The Counter Selection section allows you to select a custom subset of counters to perform a profile with. The selections made will persist across uses of GPU PerfClient and can also be saved and loaded if there are different sets of counters that you find useful.
The Per-Draw Data section of the Frame Profiler window has two main areas, a tree of "Available Counters" and a list of the currently "Selected Counters".
Here you can quickly select groups of counters to include in the Counter Selection list. Each group may contain one or more counters. You can either click on the group to toggle the group's counters on / off, or select individual counters within a group.
See the list of Available Counters for more information on the data that is accessible. Also, the Performance Analysis Suggestions can help you get started in profiling your application.
This list shows the currently selected counters that will be used during the profiling. Counters can be removed from this list by double-clicking on the counter name, or unchecking the counter from the tree of available counters. There is a limit on the maximum number of selected counters.
- Save Selection - Allows you to save the current counter selection to file for later usage.
- Load Selection - Allows you to load a saved counter selection file to quickly enable your own set of counters.
- Profile - Use this to profile the application with the selected set of counters. You must have at least one counter selected to perform a profile. this may take several seconds or up to a minute depending on your application and the number of passes required to get all of the counter data.
- Passes Required - Indicates the number of passes required to profile the selected counters. If the number of draw calls varies from one frame to the next, you should try to keep the number of passes to 1. However, if the application does not have variable draw calls, then it does not matter how many passes are required.
- In OpenGL applications, the "GPUTime" counter is accessed via the GL_EXT_timer_query / GL_ARB_timer_query extension. This extension is written such that an application cannot use more than one GL_TIME_ELAPSED queries at the same time and that generation of queries cannot occur while any type of query is active. As such, GPU PerfStudio will stop active queries as needed so that it can successfully perform its profiles. This may result in the application receiving unexpected results when it gets the data related to a timer query, occlusion query, or transform feedback query. If the change in the results affects the rendering of the application, the profile results may not be an accurate representation of the typical performance of the application.