3DMark03 Mother Nature IQ comparison b/w FX & R300
- Thread starter binmaze
- Start date
I dunno, but those are the three places PS 2.0 is used in that scene: overbright sky, overbright grass in sunlight, and water. And NV30 appears to be rendering all three incorrectly. 
(Side note: if you go to the source review they also have a "highest quality" comparison. NV30 has the same problems with sky, grass and water, but for some reason R300 has problems rendering the right frame!)
Beside the questions regarding different driver versions on NV30, I'd like to know if the effect shows up under normal settings (these are 4xMSAA/8xAF)...
(Side note: if you go to the source review they also have a "highest quality" comparison. NV30 has the same problems with sky, grass and water, but for some reason R300 has problems rendering the right frame!)
Beside the questions regarding different driver versions on NV30, I'd like to know if the effect shows up under normal settings (these are 4xMSAA/8xAF)...
THe_KELRaTH
Regular
I wouldn't be suprised that if the Nv 43.00 or 42.63 drivers were used (the ones where the GFFX scores are much lower) that this is corrected.
Personally I still think that the optimised drivers forced everything to 12bit Integer rather than 16bit FP as so far all PS 2 shader tests comparing FP16 v FP32 have had the same results.
Personally I still think that the optimised drivers forced everything to 12bit Integer rather than 16bit FP as so far all PS 2 shader tests comparing FP16 v FP32 have had the same results.
i am guessing that the difference in water is due to shader issues as well.
and here are gifs for both:
4xaa & 8xaf.
highest quality.
it looks like there is some difference with physics in the ati-hq shot, does anyone understand why?
and here are gifs for both:
4xaa & 8xaf.
highest quality.
it looks like there is some difference with physics in the ati-hq shot, does anyone understand why?
OpenGL guy
Veteran
Maybe worm can tell us. To me, it looks like an application issue (wrong seed value).kyleb said:
I used to see similar things in UT2003 when it used to use a different random seed for botmatches each time.
lol Uttar, the r300 and nv30 are right just about right on the money with eachother at 1280x10240 like in those pics, at least according to hardocp. granted that was at default settings and with aa and af at what they are in the pics; i would wadger that the r300 trounces the nv30 in scoreing.
also thanks for the backup antlers4, i was just working within my ablities.
i should mention though, they came out looking remarkably close to the original jpgs from what i saw as i was making it. i used high quality settings for conversion as you can tell by the filesize and i tend to think that most of the compression has already been done to the images and it won't get much better than they are in gif forum. however, if someone a bit more handy with html than me can make jpgs animate, i suppose it would be better for the rest of you. or better yet if someone with both cards and a licensed copy of 3dmark03 has the time, we could really see it done right.
lastly, i agree with OpenGL guy that it looks like some some seed value issues between the pics. but the branches are even misplaced, so i still wonder if phyicis is not playing a part.
also thanks for the backup antlers4, i was just working within my ablities.
i should mention though, they came out looking remarkably close to the original jpgs from what i saw as i was making it. i used high quality settings for conversion as you can tell by the filesize and i tend to think that most of the compression has already been done to the images and it won't get much better than they are in gif forum. however, if someone a bit more handy with html than me can make jpgs animate, i suppose it would be better for the rest of you. or better yet if someone with both cards and a licensed copy of 3dmark03 has the time, we could really see it done right.
lastly, i agree with OpenGL guy that it looks like some some seed value issues between the pics. but the branches are even misplaced, so i still wonder if phyicis is not playing a part.
TheMightyPuck
Newcomer
I love the comment that it is "difficult to tell who has the edge" when the 9700 clearly has better IQ in those shots.
2B-Maverick
Newcomer
I would love to know which one is the "intended result"!
Would it be possible for anyone to use the reference rasterizer for a comparision picture?
Would it be possible for anyone to use the reference rasterizer for a comparision picture?
TheMightyPuck
Newcomer
MfA said:
I think the FX is using some of it for cooling
Johnny Rotten
Regular
TheMightyPuck said:I love the comment that it is "difficult to tell who has the edge" when the 9700 clearly has better IQ in those shots.
I think you're stretching it a good bit. The IQ of those images are far far more similar than disimilar.
TheMightyPuck
Newcomer
Johnny Rotten said:TheMightyPuck said:I love the comment that it is "difficult to tell who has the edge" when the 9700 clearly has better IQ in those shots.
I think you're stretching it a good bit. The IQ of those images are far far more similar than disimilar.
It's not that big a deal to me but to specifically say it is a wash and then post pictures that seem to contract the statement seems funny. That said I wouldn't kick either card outta bed for eating crackers.
It's difficult to put one's finger on the difference between the two images..
One thing for certain though- frame 1669 is NOT the same frame between the two cards, which is odd. The blades of grass and falling leaves are in different locations, which makes it difficult to try and do exact comparisons here.
The thing that sticks out the most to me (besides the obvious color differences) are the tree leaves in the upper-right tree. On the GF-FX, the leaves are thin, sharp and smaller.. whereas they are fatter, more colorful on the ATI... but truly this could be a different perspective angle on the leaves if the frames are actually 3-4 apart from eachother, which looks like the case.
If anything, before anything objective can be arrived, FutureMark needs to comment on this, as well as clearer defined methods of reproducing these shots in an effort to hopefully reduce any possible human error if this is the case. It would seem that either:
1) The frame counter at the bottom right corner isn't accurate between different IHVs
or
2) If this is the result of some sort of "IQ" test built-in to the benchmark that captures an exact frame, there appears something faulty with it capturing the same, precise frame.
After all- differences in shaders and the like shouldnt cause a tree poly at location X,Y,Z in 3D space to be at a different location at the same rendering time!
One thing for certain though- frame 1669 is NOT the same frame between the two cards, which is odd. The blades of grass and falling leaves are in different locations, which makes it difficult to try and do exact comparisons here.
The thing that sticks out the most to me (besides the obvious color differences) are the tree leaves in the upper-right tree. On the GF-FX, the leaves are thin, sharp and smaller.. whereas they are fatter, more colorful on the ATI... but truly this could be a different perspective angle on the leaves if the frames are actually 3-4 apart from eachother, which looks like the case.
If anything, before anything objective can be arrived, FutureMark needs to comment on this, as well as clearer defined methods of reproducing these shots in an effort to hopefully reduce any possible human error if this is the case. It would seem that either:
1) The frame counter at the bottom right corner isn't accurate between different IHVs
or
2) If this is the result of some sort of "IQ" test built-in to the benchmark that captures an exact frame, there appears something faulty with it capturing the same, precise frame.
After all- differences in shaders and the like shouldnt cause a tree poly at location X,Y,Z in 3D space to be at a different location at the same rendering time!
Similar threads
