GT4 "Mother Nature" Precision

Dave Baumann

Dave Baumann

Gamerscore Wh...
Moderator
Legend
OK, I've taken a couple of quick shots in the 3DMark03 IQ tester using the 320 and 330 builds of 3DMark03 with an NV35. If you wish to repeat this test the frame I'm using is frame 200.

NV35 -> build 320
NV35_320.jpg


NV35 -> build 330
NV35_330.jpg


R350 -> build 320
R350_320.jpg


R350 -> build 330
R350_330.jpg


Even with this jpg (yes, I know, Originals are in the zip file) the precision differences appear to be there on NV35 (and probably all NV3x), even if they would be difficult to spot running. If you download the full image for NV35 you can also see the lighting on the turtle appears to be in a completely different position.

The R350 shots appear to be identical as far as I can see, which would validate ATI's claim that their shader does the same thing to FM's.

Original NV35 PNG's (1600x1200) -> http://www.beyond3d.com/previews/nvidia/nv35/iq/nv35.zip (~5MB)
Original R350 PNG's (1600x1200) -> http://www.beyond3d.com/previews/nvidia/nv35/iq/r350.zip (~5MB)
 
OK, see update. R350 comparions there now and downloads to original PNG images.

Reverend said:
It's 4AM your time Dave, for God's sakes.... go get some sleep!
Click to expand...

5 actually. And yeah, I'm gonna hit the hey now!
 
Thanks Dave, that really helps me to better understand just what the heck the cheating/optimizing difference is all about. :)

EDITED BITS: Ok, quit laughing at me already Brent....I told ya I was tired! :rolleyes: :LOL: :LOL:

Zzzzzzzzzz.
 
Any Idea why the water is at a different level on the nv35 compared to the r350? Both the before and after r350 shots have the same water level. Both the before and after nv35 shots have the same water level. But the 2 nv35 shots have a different water level than the 2 r350 shots.

Other than that, the lighting on the turtle is completely different on the nv35 with the lower precision. There are other differences throughout, but that is the most drastic.

The r350 before and after shots are exactly the same as each other. I scrolled back and forth between them at least a hundred times and I can't find one single miniscule difference.
 
I will soon be posting images that show the difference squared ( this shows off the big difference and ignores the smaller ones ) of the NV35.

I wont do this for r 3x0 because only 4 pixels had red green or blue components off by more then 1 I will however upload images showing each pixel that was off ( each component that was off will be maxed to 255 ).

Something else to note is that some of pixels in the nv35 were off by so much they went over the max value when squared!!!. ( i.e. off by 16 or more ).
 
That's the result of NVIDIA changes to pixel shaders to double GT4 performance avoiding slowdowns and stutterings?

Those NVIDIA driver guys sure knew WTF to do with quality trade-offs for performance than this bad quality trade-off.
 
Images are uploading btw if you full bright all the compents that are off the damn water is 90% coloured ( every few pixel wasn't off but its just random ) you can even clearly see the grass over the river I'm surprised that dark patch of tutrle isn't purely colour but a majority of turtle is white I'm not gonna upload the image because if you can't see where its off in the difference sqaure then your not going to notice it in the benchmark/game. Still uploading :/

Okay its done www.lexicon.net/mccann/diffs.zip
 
You must log in or register to reply here.
Back
Top