OpenGL ES 2.0: Efficient Rendering of Static and Dynamic Vertex Data

2
votes

I am writing an iOS/Android game and looking for the most performant way to render my vertex data with OpenGL ES 2.0. I have two different kinds of data: dynamic data that changes its attributes every frame, for example the player or animated background objects, and static data such as the static background or the terrain. I googled a lot since yesterday, but I could not find a clear and unique answer to the question of what is the best was to render such data.

There are basically three options for rendering such data (If I do not miss one. If so, feel free to correct me.):

Vertex Arrays Only: Just fill your vertex every frame on the CPU (including the dynamic data).

Vertex Buffer Objects Only: Allocate a VBO on the GPU with GL_DYNAMIC_DRAW where both, the dynamic and static data is stored. The dynamic data is then updated every frame via glBufferSubData.

Use both: Static data is stored and render with a VBO and the dynamic data is rendered with a Vertex Array. With this option, we need two rendering passes, one for rendering the VBO and one for rendering the vertex array.

Since the first option does not exploit the immutability of the static data and since the third option requires two rendering passes, my guess is that I should go with the second option. However, I am absolutely not sure about this and I hope you can clarify my confusion.

1
graphicsopengl-esvbovertex-array
AFAIK both main and video memory are physically the same in phones. I've only ever used android before, but I saw no difference in vertex array (dynamic) and VBO (buffering/more-static) speed. Graphics can be really fiddly and in many cases unintuitive, so spending the extra time to write some simple tests can pay off. If you can, get the GPU to do the animation and changing of your data :) – jozxyqk

1 Answers

1
votes

Allocate two Vertex Buffer Objects. One with hint GL_DYNAMIC_DRAW that will be updated frequently. Allocate a second VBO for immutable data and use the hint GL_STATIC_DRAW. According to the API documentation, GL_STATIC_DRAW should be used for data that "will be modified once and used many times"; just what you need.

Speaking of two rendering passes here is probably a misuse of the term: what you do is to render your scene in two separate drawing commands. Since drawing commands run asynchronously, you should not expericence any performance hit by doing so. A second rendering pass, on the other hand, is when you render the entire scene twice (see for example here) with different settings, or when you do some image processing effects on outputs of previous rendering passes.