PC polygon performance
- Thread starter Anonymous
- Start date
Another thing to consider about Ut2k3 vrs doom3. Is that the new unreal engine is available now to licensee were as doom3 is not at that status yet. During the summit Tim S even stated that his new engine was not designed to have the IQ features of doom3. He chose a different path and I have to agree that for his market, its not a bad call.
I doubt there are many folks who have been disappointed with the various DOOM3 shots so far. But I agree - it looks incredible in motion.
Insofar as screenshots are concerned, per pixel lighting can't be seen. What impresses me the most when looking at screenshots are the colors and sharpness.
Actually, you can calculate shadow volumes with vertex programs, just project every back-face vertex at a finite fixed distance. This is not the exact silhouette (has mush more polygons) but works with closed shapes and on-board generated geometry (like N-Patches). Also, there are more robust techniques that work with “open†shapes, but you have to change your geometry. I think I saw a couple of ATI/NVIDIA docs outlining the details.
Carmack said it was a bit slower on a GF3, but the CPU usage was low (about 10%).
As for shadow algorithms, I think the “Deep Shadow Maps†by Lokovic and Veach is currently the best practical algorithm. Ray casting is of course the most accurate and simple but is not yet practical if you don’t want just simple sharp opaque shadows.
Carmack said it was a bit slower on a GF3, but the CPU usage was low (about 10%).
As for shadow algorithms, I think the “Deep Shadow Maps†by Lokovic and Veach is currently the best practical algorithm. Ray casting is of course the most accurate and simple but is not yet practical if you don’t want just simple sharp opaque shadows.
Pavlos said:
This won't work on hard edges (where neighbouring faces are not in the same smoothing group). You can create a fully smoothed version of the object, but doubt it would look anywhere near correct.
(Of 'cos N-patches don't like hard edges, but thats a different story.)
Yeah, thats the right thing - but you cannot do them with vertex shaders.
He also said that 8x Aniso has less than 30% hit on the GF3
You are correct, the simple algorithm (pushing back-facing vertices to “FARâ€) works only for closed, highly and smoothly tessellated objects, but the vertex program is short and the geometry of some games satisfies the conditions.
You can add a degenerate quad at every edge and the algorithm will work for everything, even for a single triangle.
