Unfortunately for the moment the demo can't give this number. The fact is that mandelbrot isn't an app to show us what kind of quality difference whe can have in a game with partial precision vs full precision but what is the actual precision of X card with Y driver.
What i can said is that the precision from what i see is that :
- 44.03 Detos force partial precision for all case on FX5600/5800 in D3D
- 44.10 Detos use full precision by default and, when you ask it, partial precision in D3D
- 44.03 & 44.10 Detos use full precision in OGL ARB Fragment program
extension.
- Catalyst 3.4 doesn't use partial precision in D3D when you ask for it, it's always full precision
With mandelbrot we can see that GFX partial precision < ATI partial/full precision < GFX full precision from a quality point of view (well it's quite logical to have FP16 (or FX12?) <FP24<FP32
)
I haven't checked the 5900 with mandelbrot demo and now the card is gone ... but in the teapot ATI demo the 5900 just look as fine as 5600/5800 with 44.10 even with 44.03, so i think there is no precision problem /w 44.03 on 5900 in D3D.
What i can said is that the precision from what i see is that :
- 44.03 Detos force partial precision for all case on FX5600/5800 in D3D
- 44.10 Detos use full precision by default and, when you ask it, partial precision in D3D
- 44.03 & 44.10 Detos use full precision in OGL ARB Fragment program
extension.
- Catalyst 3.4 doesn't use partial precision in D3D when you ask for it, it's always full precision
With mandelbrot we can see that GFX partial precision < ATI partial/full precision < GFX full precision from a quality point of view (well it's quite logical to have FP16 (or FX12?) <FP24<FP32
)I haven't checked the 5900 with mandelbrot demo and now the card is gone ... but in the teapot ATI demo the 5900 just look as fine as 5600/5800 with 44.10 even with 44.03, so i think there is no precision problem /w 44.03 on 5900 in D3D.