[WMCoolmon]

I've heard that the FX5200 runs Halo worse than the MX440 precisely because it supports pixel shaders, and has to do the calculations for version 2 instead of 1.

Edit: What would really help give a sense of scale would be motion blur.

[Flipside]

Yep, I run an overclocked 5200 in the other machine, it only mildly outperforms the old fanless Radeon 9200 I had in there before it, and this things got something akin to a Jumbo Jet intake strapped to it

Some games, like X will work fine with bump-mapping, but hates shadows. Doom 3 is good with shadows, but really doesn't like normal maps. Sims 2, however, seems to handle bump mapping, normal mapping, etc etc without any hint of a slowdown. I've yet to find any common denominator to what works and what doesn't.

[Unknown Target]

I think it'd be great if we could simply have an auto-LOD feature that automatically removes polies as the ship goes farther away. It'd look better and more importantly, alleviate the modeler workload.

[WMCoolmon]

Unless the video card has onboard support for such a feature, dynamically doing it would actually cause slowdowns.

If LODs were generated when the ibx were made, it'd look worse, but would require coming up with a poly-reduction algorithm.

Edit: Because I feel like explaining a little better.

The whole vertex buffer upgrade thing took advantage of the ability of video cards to store model geometry on the on-chip RAM, much faster than sending it over the AGP bus every frame. Doing on-the-fly LOD calculations would mean that the model would have to be re-sent over the AGP bus, taking up bandwidth that could be used for textures and stuff, possibly forcing the video card to wait on the model. The end result would be that the effects of the vertex buffer upgrade would be nullified, or reduced, depending on how often a lower-poly model was generated and sent. Not to mention the time the CPU would spend calculating it all out.