Rendering Multiple Models

It depends on many factors. When you say you're drawing models, are you talking about things like player models or potions on the ground, or are you calling each tile on the ground a model? If your question lies in the realm of tile-based optimizations, then unless you're drawing large tiles with limited view (i.e. 3rd person straight down), you'll probably need to optimize the tiles from the start...that is draw more than one at a time. To optimize, you need to group them according to texture, add the first group to a vertex buffer or index buffer, draw that group, and move to the next group. There are many optimizations you can place in that as well. For instance, you could add entire columns and rows to the buffers at a time, preventing having to re-calculate the entire terrain in view.

If you're asking about character models and item models, houses, whatever, then since you're new at Direct3D, I'd recommend simply setting the world matrix and drawing them. You won't be drawing that many per frame most likely, at least if you have some kind of scene organization such as a Quadtree. Get that part working, then move on to geometry instancing and other optimizations later on. It will be more fun to get a working game engine that does minimal stuff than spending so much time on one part that it takes you years to get an entire engine together, and at some point you'll get bored with it.

I recommend making an outline, and define versions of the engine and what it will support, then work towards getting each version complete. Learn about optimizations ahead of time and try your best to keep them in the back of your mind as you're building the engine, so as not to design it in a way that will make optimization complicated later. Put these down on your version outline.

The optimal approach is to get a nice balance between CPU and GPU. While the GPU is drawing several models in a single vertex buffer, you're sorting and building several more to be drawn next. This way, you utilize the GPU and CPU at the same time.

Lastly, don't forget to use static vertices when you can. For example, tiles are going to always be in the same location. Once you lock the vertex buffer and update the position of the tile, you don't have to do it again. Same goes for system memory. Once you create the tile in system memory, all you have to do is copy it to video memory when it is ready to display. I do this by searching a quadtree for which tiles are currently visible, then add them to a vertex buffer for each texture. Don't load a model and call the same SetTransform world transform each frame if it will never move. Update the verts once and leave them. Hope this gives you some ideas and isn't too far off on a tangent.

Thanks, that's all great feedback.

Geoff