Reuse VertexBuffer or new VertexBuffer object?

Question

I'm trying to render bitmap fonts in directX10 at the moment, and I want to do this as efficiently as possible. I'm having a hard time getting a start on my design because of this question though.

So should I reuse a single VertexBuffer or make multiple VertexBuffer objects?

Currently I allocate one dynamic VertexBuffer per Quad object in my program. This way I wouldn't have to map/unmap a VertexBuffer if nothing moves on my screen. For fonts I can implement a similar method on where I allocate one buffer per text box, or something similar.

After searching I read about reusing a single VertexBuffer for all objects. Vertex caching came up also. What is the advantage/disadvantage of this, and is it faster than my previous method?

For last, is there any other method I should look into rendering many 2d quads in the screen?

Thank you in advance.

Answer 1

You may reuse VertexBuffer objects but this is harder to implement, but as @SeanMiddleditch said in his comment, this shouldn't be your goal, and creating new VBOs shouldn't matter in your application.

Currently I allocate one dynamic VertexBuffer per Quad object in my program

Per quad is an overkill and is probably the worst. This way your practically emulating immediate mode which has been deprecated for good.

Also storing all your scene in one vertex buffer is usually a bad idea. You may need to cull objects and not draw them in the first place in many occasions. Storing them in one VBO will particularly prevent you from doing culling operations.

Your best bet is to create new vertex buffers for each 3D model, try not to store small data or huge data.

My recommendation is that it shouldn't be your concern unless your application really took a performance hit. In that case profile and optimize accordingly.

After searching I read about reusing a single VertexBuffer for all objects. Vertex caching came up also. What is the advantage/disadvantage of this, and is it faster than my previous method?

Vertex caching is an advanced optimization and shouldn't be done unless you have a very performance critical application (like the next big AAA game, or a scientific visualization that deals with millions of data chunks), and btw vertex caching shouldn't matter for smaller scenes.