2 replies to this topic

[mrheisenberg]

Would generating a command list from multiple threads with deferred contexts have any noticeable gain on performance with a deferred renderer?I mean you aren't changing shaders for every type of material and instead you use a branch or some other way of determining the material inside the lighting shader and shader binding is the heavies state change,right?

[Hodgman]

If you've got 50,000 draw calls, it still might be using a decent amount of CPU cycles.

Put a timer around the part of your code that actually makes D3D calls. If it's too large, then it's worth optimizing.

[MJP]

Who says you won't change shaders when you change materials? That would be assuming that every single material will fill the G-Buffer in the exact same way. As a really simple example, you might have a shader that does normal mapping and one that doesn't.