Performance Profiler

A performance profiler displays performance data in a timeline. It reports how much time is spent per each frame for updating all aspects of your project: rendering nodes that are in view, updating their states, executing scripts with game logic, calculating physics, etc.

With the profiler, you can:

Detect the bottlenecks of your application
Check if art assets optimization is required
Check if code optimization is required
Compare the profiling results before and after the changes

To turn the profiler on, click Tools -> Performance Profiler and choose the required profiling mode:

Notice

You can also set a hot key (or a key combination) that will run the profiler in the Hotkeys section of the Settings window. However, you will have to create a custom preset. Then you can press this hot key for several times to cycle through profiling modes.

The following profiling modes are available:

Generic profiler shows only the general statistics block.
Rendering profiler shows the detailed rendering statistics and the timeline chart.
Physics profiler shows the detailed physics-related statistics (within the Physics radius) and the timeline chart.
World Management profiler shows the statistics on the whole loaded world.
Thread profiler shows the statistics on loading threaded resources.

To show profiler statistics in the in-game modetype show_profiler command and a value from 1 to 5 in the console. To disable the profiler in the in-game mode, type show_profiler 0 in the console.

Generic Profiler Enabled

Total	Total time in milliseconds taken to both calculate and render the current frame. This is the duration of the main loop in the application execution sequence. Total = Total CPU + Waiting GPU
Total CPU	Total time in milliseconds taken to prepare the current frame (including update, render, and swap).
Total GPU	Total time in milliseconds taken to render the current frame on the GPU. Notice This counter may not work on some GPUs.
Update	Time taken to update application logic. This includes executing all steps in the *update()* function of the world script. It also includes the update of states of all nodes (for example, update of the skinned animation or of a particle system to spawn new particles). If the Update time is too high, it signals that you need to optimize the application logic executed each frame. You may also need to decrease the number of objects in the world, as updating their states (spawn particles by the particle systems, play skinned mesh animation, etc.) increases the load.
Render CPU	Time taken to prepare all data to be rendered in the current frame and feed rendering commands from the CPU to the GPU. If Render CPU time is too high, it signals that art assets may need to be optimized, for example: LOD system should be used; polygon count of models should be reduced.
Waiting GPU	Time between completing all calculations on the CPU up to the moment when the GPU has finished rendering the frame. (See the illustration). This counter is useful to analyze the bottleneck in your application's performance. When Waiting GPU time is equal to 0, it signals that scripts take too long to be updated and calculations are too intensive for CPU to perform them fast enough. In this case, your application has a CPU bottleneck. Optimize your update block in the world scripts or reduce the number of objects updated each frame. High Waiting GPU time means one of the following: low framerate signals that there exists a GPU bottleneck. The art content needs to be optimized in this case. consistently high framerate means you have free CPU resources available to process more numbers in the update() of the world script.
Interface	Time taken to render all GUI widgets.
Physics	Time taken to perform all physics calculations.
Heap	Size of all memory pools allocated for the application. Unigine allocator allocates memory in pools which allows the allocation to be faster and more efficient (if USE_MEMORY directive is used, by default). As the memory is allocated in pools, the counter value increases stepwise.
Memory	Size of all memory blocks allocated on demand. This counter reports how much memory in allocated pools application's resources really use. If Memory size steadily increases, it may signal that there are memory leaks. Check if all created objects and variables that are no longer used are properly deleted.
System	Size of RAM memory used for the application.
Allocations	Number of allocation calls during the frame. (This counter reports an allocation call even if several bytes are requested to be allocated).

Rendering Profiler Enabled

The following statistics is displayed in addition to the generic one:

Sounds	Memory amount, currently used for sound samples, in megabytes.
Meshes Limit	Maximum amount of VRAM memory that can be used for mesh geometry, in megabytes.
Meshes	VRAM m amount currently used for mesh geometry, in megabytes.
Textures Limit	Maximum amount of VRAM memory that can be used for textures, in megabytes.
Textures	VRAM m amount currently used for textures, in megabytes.
Textures Cache	VRAM memory amount currently used for textures cache, in megabytes.
Buffers Render	VRAM memory amount currently used for rendering buffers (Gbuffer, post-effects, etc.), in megabytes.
Buffers Shadows	VRAM memory amount currently used for shadows maps, in megabytes.
Async Buffer	Memory amount currently used for the async buffer, in megabytes. Intermediate buffer to which resources are asynchronously loaded during the streaming process. Notice Available for OpenGL only.
Async Buffer Indices	Memory amount currently used for mesh indices buffer, in megabytes. This buffer is used to store indices of meshes asynchronously loaded during the streaming process. Notice Available for OpenGL only.
Grasses	VRAM memory amount currently used for grass nodes, in megabytes.
Terrains	VRAM memory amount currently used for terrains, in megabytes.
Allocations Textures	Number of memory allocations for textures during the frame.
Compile Shaders	Number of shaders compiled during the frame.
Dynamic Reflections	The number of dynamic reflections drawn per frame. In case of cubemap reflections, if all six faces are updated, six reflections are rendered each frame.
Lights	Number of light passes rendered per frame. This means that the counter displays the number of all light sources that are currently seen illuminating something in the viewport. This value also includes additional passes for rendering lights in the reflecting surfaces (if dynamical reflections are used). Plain 2D reflection will multiply the number of rendering passes by two, while cubemap-based reflection with six faces updated each frame will multiply the number of rendering passes by six. Notice Each light redraws mesh polygons it illuminates. That is why the higher the number of light sources, the higher the number of polygons the graphics card has to render, and the lower the performance. For example, using two omni lights will as much as double the rendered geometry they shine on.
Shadows	Number of shadow passes rendered per frame. Each light requires a shadow pass to calculate the shadows. Again, if there are reflecting surfaces with shadows drawn reflected, this will increase the number of shadow passes.
Decals	Number of decals rendered per frame (in all rendering passes).
Surfaces	Number of surfaces rendered per frame (in all rendering passes). Each light source doubles the number of surfaces if they are lit.
Triangles All	Total number of triangles rendered per frameincluding all polygons that are currently visible in the viewport as well as the ones rendered in the process of shadows rendering.
Triangles Shadows	Number of triangles rendered per frame in the process of shadows rendering. Each light source has to redraw the geometry it illuminates, increasing the overall count of rendered triangles. In order to avoid GPU bottleneck, keep the number of dynamic light sources and their radius as low, as possible.
Triangles Viewport	Number of triangles rendered per frame. This includes all polygons that are currently visible in the viewport (geometry).
Primitives	Number of geometric primitives rendered per frame. This includes points, lines, triangles, and polygons. The visualizer and the profiler itself also add to this counter. The value differs dramatically if tessellation is used. In this case, Triangles reports the number of triangles in the coarse mesh, while Primitives shows statistics on the number of tessellated primitives.
Dips	The number of draw calls. The higher the number of identical mesh surfaces with the same material, the more effective the instancing is (enabled by default). This means, the number of draw calls is minimized offloading both the CPU and the GPU. You can compare the number of surfaces ( Surfaces) and the number of DIPs used to render them. For example, if there are 30000 surfaces and 1000 DIPs, it means that 30 instanced surfaces of meshes are rendered per only one draw call (Surfaces/Dips). Thus the instancing provides performance boost.
Shaders	Number of shaders set per frame. (Shaders are set in each of the rendering passes; hence if only one material used, its shader still needs to be set several times. When nothing is visible and the screen is black, even in this case the composite shader is still used.)
Materials	Number of materials set per frame. (Materials are set in each of the rendering passes.)

Notice

Values less then 1 Mb are displayed as 0.

This profiler shows statistics within the Physics radius.

Physics Profiler Enabled

Notice

To show valid information on

PUpdate
PResponse
PIntegrate

the physics simulation should be run in the single-threaded mode ( physics_threaded console command should be set to 0 or 1).

The following statistics is displayed in addition to the generic one:

PIslands	Number of physical islands within the physics radius that could be calculated separately. The lower this number, the less efficient multi-threading is, if enabled.
PBodies	Number of bodies within the physics radius.
PJoints	Number of joints within the physics radius.
PContacts	Total number of contacts within the physics radius; this includes contacts between the bodies (their shapes) and body-mesh contacts.
PBroad	Duration of the broad phase of physic simulation when potentially colliding objects are found.
PNarrow	Duration of the narrow phase when exact collision tests are performed.
PUpdate	Duration of the update phase when objects are prepared for their collision response to be calculated.
PResponse	Duration of the response phase when collision response is calculated and joints are solved.
PIntegrate	Duration of the integrate phase when physics simulation results are applied to bodies.
PSimulation	Duration of all simulation phases added together.

Watch an overview of the Physics Profiler options in our video tutorial on physics.

This profiler shows statistics on the whole world.

NoticeWhen in the editor mode, enabling and disabling of the node, body or a joint can increase values of world profiler counters, as a (limited) number of clones are created for undo/redo purposes.

World Management Profiler Enabled

The following statistics is displayed in addition to the generic one:

WNodes	Total number of nodes in the world (both enabled and disabled).
WBodies	Total number of bodies in the world.
WJoints	Total number of joints in the world.
WSpawn	Time in milliseconds that the engine spends on generating content in procedural nodes (such as grass, clutters, world layers).

Thread Profiler Enabled

The following statistics is displayed in addition to the generic one:

AsyncQueue	Time of asynchronous loading of resources (files/meshes/images/nodes), in milliseconds.
Sound	Time of asynchronous loading of sounds, in milliseconds.
PathFind	Time of asynchronous pathfinding calculations, in milliseconds.

In This Article:

Activate Profilers#

Generic Profiler#

Rendering Profiler#

Physics Profiler#

World Management Profiler#

Thread Profiler#