How do you speed up your real-time rendering?

1 Use LODs and culling

LODs, or levels of detail, are simplified versions of your 3D models that have fewer polygons, textures, and other details. You can use LODs to reduce the amount of geometry that you need to render for objects that are far away from the camera or not visible at all. Culling is the technique of discarding objects that are outside the view frustum, occluded by other objects, or too small to be noticeable. By using LODs and culling, you can save a lot of rendering time and memory by only drawing what is necessary for the current frame.

2 Optimize your shaders and materials

Shaders are programs that run on the GPU and determine how your objects look on the screen. They can create various effects, such as lighting, shadows, reflections, and textures. However, shaders can also be expensive to execute, especially if they involve complex calculations, loops, or branching. To optimize your shaders, you should try to minimize the number of instructions, avoid unnecessary computations, and use constants and uniforms instead of variables. Materials are collections of shaders and parameters that define the appearance of your objects. To optimize your materials, you should try to reuse them as much as possible, avoid overusing transparency and blending, and use texture atlases and compression to reduce the size and number of textures.

3 Use batching and instancing

Batching and instancing are techniques that allow you to render multiple objects with the same material and geometry in one draw call. A draw call is a command that sends data from the CPU to the GPU for rendering. Each draw call has some overhead, such as setting up the state, validating the parameters, and transferring the data. By reducing the number of draw calls, you can improve the CPU-GPU communication and reduce the driver overhead. Batching is the process of combining multiple objects into one mesh, while instancing is the process of rendering multiple copies of the same object with different transformations. Both techniques can help you render large numbers of similar objects, such as trees, rocks, or characters, more efficiently.

4 Use occlusion queries and depth pre-pass

Occlusion queries and depth pre-pass are techniques that can help you avoid rendering objects that are hidden by other objects. Occlusion queries are queries that you can send to the GPU to check whether an object is visible or not before rendering it. For example, you can use occlusion queries to test if a large building is blocking the view of a smaller object behind it. If the query returns false, you can skip rendering the smaller object and save some GPU time. Depth pre-pass is a technique that involves rendering your scene twice, once with only the depth information and once with the full shading. The depth pre-pass can help you avoid overdraw, which is the situation where you render the same pixel multiple times with different objects. By using the depth buffer from the first pass, you can discard pixels that are behind other objects and only render the visible ones.

5 Use post-processing effects wisely

Post-processing effects are effects that you apply to the final image after rendering the scene, such as bloom, motion blur, color correction, and anti-aliasing. Post-processing effects can enhance the visual quality and realism of your graphics, but they can also be costly to perform, especially if they involve multiple render targets, full-screen passes, or complex filters. To use post-processing effects wisely, you should try to limit the number and intensity of the effects, use lower resolutions or downsampling, and use temporal techniques that reuse information from previous frames. You should also profile your performance and identify the most expensive effects and adjust them accordingly.

6 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?