Why 3Dc may not always be a better solution

[Scali]

Sounds like SM3.0 ( in its current incarnation) vs. SM 2.x, so would you argue that it is premature to utilize SM3.0?

No. That is a completely different story. You can simply recompile SM2.x shaders to SM3.0 without changing any code. If this improves performance, I don't see why you wouldn't use this option.

One may argue however that there's not enough hardware around to write specific SM3.0 code for. As far as I know, Futuremark also didn't write specific SM3.0 code, but simply compiles their code for the best model the hardware supports.

At any rate, there are no disadvantages to using SM3.0. But 3Dc does have some disadvantages, and requires both code and artwork to be modified.

In short, the analogy is completely superficial.

 

[Dio]

Why does it require artwork to be modified? It's just a different way of compressing the data, and I suspect that it would be trivial to write a 3Dc -> DXT5 conversion utility.

 

[Patric Ojala]

Scali's original post in this already fairly long thread has many good points, but it is not true that ATI would be quiet about using DXT5 for compressing normal maps. ATI's guys actually offered the most valuable help implementing DXT5 normal map compression.

3DMark05 does not include 3Dc because we already use DXT5 for normal map compression. All hardware capable of running 3DMark05 support DXT5, the implementation is just as elegant and DXT5 is a DX feature, not even new for DX9, and not IHV specific. So DXT5 normal maps would have been needed anyway, but adding a set of 3Dc normal maps would have added probably 100-150MB more to the download package. It’s not just a few added lines of code, you would need all normal maps included also in 3Dc format. You can of course start playing with texture conversions, but we did that in an earlier 3DMark version and have no desire to go there again.

... and yes, DST/PCF does offer a performance increase. We have no info on 3Dc increasing the performance over our DXT5 implementation.