Quake shown by ATI for new mobile 2300 graphics

[ToTTenTranz]

[ToTTenTranz]

[spiral]

not a sensitive spot, just a dead thread.

[ToTTenTranz]

[James S]

[ToTTenTranz]

[Dan East]

I can't comment on any 2300 specs at that level, nor can I compare it to some other chip. ATI played a dominate role in creating the OpenGL ES spec. They even drafted it in the first place. Imagination is clearly playing catch-up in this area, as they only announced at CES 2004 they would have OGLES support later this year, while the 2300 already has complete (and demonstrated) OGLES drivers in place.

Although PowerVR MBX may hit the market first, ATI obviously kept them honest, and prevented Imagination from sliding by and forcing some proprietary 3D api (probably derived from the Dreamcast) on us.

Finally, as I've said before, no XScale PPC can currently make full use of the 2300. It cannot push polys at it fast enough, even with geometry caching. So there isn't much point in the PowerVR MBX being any faster. The rendering takes place in parallel to the CPU anyway, so there would be no net gain. However the bigger issue is pure rendering quality, and driver robustness. ATI at least has Imagination beat with the latter. ATI played a role describing the specs, and thus PowerVR MBX is likely not as well suited for OpenGL ES (or more accurately, OpenGL ES is likely not as well suited for PowerVR MBX). ATI may also have them beat on the former too, but that subjective comparison would have to wait until PowerVR MBX has a real 100% port behind it, like GLQuake (the Tomb Raider that is running on PowerVR MBX is the basterdized PS1 version, so it doesn't count).

Dan East

[ToTTenTranz]

[Dan East]

[ToTTenTranz]

[sponge]

[Winter]

[Alexander]

[Phoenix]

[Dan East]

Oh, there is no comparison to using the 2300 and software rendering. The rendering quality of the 2300 is better in every way. Internally it uses 64 bit floating point math, so all vertex transformation and projection is extremely accurate. Seams and popping of edges I see in Varium (software) go away when using OpenGL ES / 2300. Bilinear filtering can be used in all cases with the 2300, which again leads to much, much better rendering quality.

There is also no comparison to performance. You can still render many, many times more polys with the 2300 than you could ever hope with software.

The argument isn't whether or not the 2300 is substantially better than what we have now (it is). The discussion is regarding the platform (XScale 400 MHz) not being able to push the 2300 to its limits. The problem is that people are wanting to compare the 2300 to the current generation of cards on desktop machines.

Regarding the lack of a FPU - that is a non-issue. The accuracy and range of floating point is massive overkill when talking about vertex positions. A range of +-32768 (integer) is more than enough for almost all scenerios. The only issue with not having a FPU is really for software engines, which are forced to use fixed point math, lookup tables, fast integer division, etc, which introduces accuracy problems. None of that is an issue with the 2300 because internally it does all of that math in floating point at the hardware level.

Dan East