I've been working on some larger projects and have been wondering about everyone's workflow. For me, things seem fine and dandy until I get up to about a million verts and then the computer starts to bog down a bit (AMD X2 4200+, 4gb, SLI 7950's) I can usually switch over from shaded to wireframe and get a bit more before it starts to slow down. How do I get several million verts to perform and render in a reasonable fashion? I usually render these out using HDR lighting and it slows me down even more in the renders....no render farm here.
there is no real magic bullet here. The only way to handle such large datasets is to use a 64bit app, and lots of ram, and a fast video card. Rendering times are always gonna be slow with HDRI, just comes with the territory. You can try breaking your scene up in passes. While that won't speed up rendering time on the whole (cause you'll be rendering multiple passes), you will at least get more immediate feedback than waiting 8 mins a frame on a render. Also, as you work, you might break your scene in display layers, so you can hide sections you don't need to see, which will also speed the process up.
The question you have to ask yourself when interacting with your scene is "Do I really need to see all 1 million+ verts together at one time?" Usually the answer is no. I tend to go through my scene and set up display layers that let me quickly and easily turn large chunks of geometry on and off when I'm working in an area that doesn't require me to see everything. We go through great lengths to set up our pipeline and workflow so that shots are broken up into pieces that are manageable. You only need to see x, y and z for animation. Layout and camera work needs to see a,b,c and standin versions of x, y, z. Etc.
I don't think you need 64 bit architechture to get around a million+ vertex scene. It can help, but it wont process much faster, and will use more memory actually.
As for rendering sometimes you just have to wait the wait. Yes you can, and probably should break your scene into render layers. If something changes hopefully the whole frame doesn't need to be rerendered. You only have to rerender the characters, their shadows and possibly any bounce they inflict on the scene. But usually the environment affects the lighting of the characters, and the characters rarely influence the lighting of the environment, unless they're holding lightsources in some way.
Layers are the way to reduce overhead in any pipeline