Doom3: Will it have a separate SM3 codepath ?
- Thread starter rms
- Start date
Sometimes he is maybe a little over optimistic 
But GF was never intended to run Doom III well : it offers just the minimum features to run the game with all the bells & whistles. I remembered an interview on VoodooExtreme at the end of 2000 and he said that the GF2 Ultra (which was the best card out there at that time) wouldn't be able to run the game well. I was shocked even if he added that the game wouldn't be available before a long time (didn't think it would take THAT long) of course today nobody cares
But GF was never intended to run Doom III well : it offers just the minimum features to run the game with all the bells & whistles. I remembered an interview on VoodooExtreme at the end of 2000 and he said that the GF2 Ultra (which was the best card out there at that time) wouldn't be able to run the game well. I was shocked even if he added that the game wouldn't be available before a long time (didn't think it would take THAT long
) of course today nobody cares What got me wondering lately is, now that all DX7 class cards will run ARB1 and all from the DX9 class (and above) use ARB2 - what about GF3/4 and the 8500? Will there be separate NV20 and R200 paths, as we were told before, or perhaps JC has extended the "less is better" approach to the DX8.x class too, and all of them will run PS1.1? 
anaqer said:What got me wondering lately is, now that all DX7 class cards will run ARB1 and all from the DX9 class (and above) use ARB2 - what about GF3/4 and the 8500? Will there be separate NV20 and R200 paths, as we were told before, or perhaps JC has extended the "less is better" approach to the DX8.x class too, and all of them will run PS1.1?
the ARB path is only for the original Radeon and it doesn't offer specular highlights. The only code path that has disappeared is the NV30 one since it's no longer necessary now that GF FX can obtain the same level of performance with the ARB2 path (for whatever reasons, this is not the topic
). On the contrary NV10, NV20, R200 are still needed because there is no ARB alternative for this level of feature. The philosophy of Carmack is simple : if there is an ARB extension he uses it. During Doom III development he replaced EXT_vertex_shader and NV_vertex_program with ARB_vertex_program. And he did the same with VBO. With this principle the NV30 code path shouldn't have existed right from the start.
I seriously doubt there will be any ATI optimizations in Doom3 as Carmack announced he was replacing the shaders in the ARB2 pathway with Cg compiled shaders (i.e. an automatic performance hit for ATI). This was done so he could get rid if the nVidia specific pathway. I can't think of a single other reason for this move than to slow down ATI cards and keep people from benchmarking/comparing the different pathways.
By the way, this is the exact opposite of what Valve is doing in Half-Life2. Valve is trying to increase performance for nVidia cards by making special dumbed-down shaders while Carmack is making ATI shaders run slower by using Cg compiled shaders. One company is trying to increase performance for all cards while the other is only increasing performance for one company.
By the way, this is the exact opposite of what Valve is doing in Half-Life2. Valve is trying to increase performance for nVidia cards by making special dumbed-down shaders while Carmack is making ATI shaders run slower by using Cg compiled shaders. One company is trying to increase performance for all cards while the other is only increasing performance for one company.
Cg is currently no longer being used either in Doom 3. Support for it is there though.
http://www.forumplanet.com/planetdoom/topic.asp?fid=5733&tid=1413239
Also Carmack is using various special ATI features to make use of abilities with the R200 (which Cg could only be used to take advantage of the Vertex Shaders on the R200 and not what pixel capabilities it has).
Robert Duffy said:
http://www.forumplanet.com/planetdoom/topic.asp?fid=5733&tid=1413239
Also Carmack is using various special ATI features to make use of abilities with the R200 (which Cg could only be used to take advantage of the Vertex Shaders on the R200 and not what pixel capabilities it has).
Well, I'm not sure that this is the case on the NV30. With only one floating point unit per pipe, there's absolutely no chance. Some performance might be returned as a result of keeping the register count down, but the lack of floating point processing power is very significant indeed. The NV35, however, does have a good chance of being faster than the R300 and perhaps at least as fast as the R350/360, especially when you factor in its depth/stencil optimizations.ChrisW said:
ChrisW said:
You're mistaken. What JC said was that he had put support for Cg in. He also said that the nv30 was using fp16 in the ARB2 path. He also said nVidia was probably "cheating" (his own words) by replacing shaders in the driver because if he changed the DOOM shaders slightly and the nv30 got big performance hits.
I don't know how from of all the above you get the idea he's out to hurt ATi card owners.
I was basing that on other quotes. I didn't see the forumplanet quote until just now. I'm not the only person to interpret those quotes that way.Mordenkainen said:
http://www.beyond3d.com/forum/viewtopic.php?topic=12006&forum=9
Similar threads
