Mandleprot Precision Outputs

[Dave Baumann]

Do we have a Mandleprot program with will show the differences between all the precisions available? i.e FXxx, FP16, FP24 and FP32?

Cheers

 

[Xmas]

You mean, a software Mandelbrot renderer emulating FP16/FP24 to compare to PS-rendered images? I don't think anyone has written such a thing yet as it makes no sense for any but that purpose.
Hey, but I know a Mandelbrot renderer with FP224 precision

 

[Dave Baumann]

I just would like comparison shots of Int and the three FP modes available currently.

 

[BoddoZerg]

You mean, a software Mandelbrot renderer emulating FP16/FP24 to compare to PS-rendered images? I don't think anyone has written such a thing yet as it makes no sense for any but that purpose.
Hey, but I know a Mandelbrot renderer with FP224 precision

http://spanky.triumf.ca/www/fractint/dz.html

 

[Xmas]

http://spanky.triumf.ca/www/fractint/dz.html[/img]

This one doesn't use arbitrary precision, but it's *fast*:
http://home.t-online.de/home/zsack/mandelbrot.html

 

[antlers]

If anyone has the hardware, the software can be found on Humus' site.

It will run FP24 on ATI cards, FP16 on NV31-34 (I think), FP32 on NV35 (?)

You need to standardize on a zoom point, probably one where artifacts start to become obvious at FP24. FP32 will look better and FP16 worse.

 

[Arun]

Yes, but the problem is that we don't know if the NV3x REALLY does that, and it's hard to tell without knowing what it's *supposed* to look like.

If anyone has the hardware, the software can be found on Humus' site.

It will run FP24 on ATI cards, FP16 on NV31-34 (I think), FP32 on NV35 (?)

You need to standardize on a zoom point, probably one where artifacts start to become obvious at FP24. FP32 will look better and FP16 worse.

 

[KimB]

Yes, but the problem is that we don't know if the NV3x REALLY does that, and it's hard to tell without knowing what it's *supposed* to look like.

Why? With a fractal it should be obvious.

If it looks better than the R3xx when zoomed, then it's got to be FP32.

If it looks worse, then it's most likely FP16, though it would be interesting to see the differences between FP16 and FX12 in one of these mandelbrot shaders.

Personally, I don't see the mandelbrot shaders as a good way of seeing what differences exist, but instead are a better way to just see what precision is actually used in the calculations. These shaders aren't going to be similar to those seen in real games. They're the ultimate in synthetic testing.

 

[Xmas]

I think with FX12 it would look blocky even without any zoom.

 

[KimB]

I think with FX12 it would look blocky even without any zoom.

I would tend to think so. But I'm not sure how the mandelbrot shader depends on precision: is it more dependent on the mantissa, or the dynamic range?

Since essentially all mandelbrot shaders use FP, i would imagine the quality is largely dependent on dynamic range, so that FX12 would look far worse than FP16.

 

[Ostsol]

I think with FX12 it would look blocky even without any zoom.

I would tend to think so. But I'm not sure how the mandelbrot shader depends on precision: is it more dependent on the mantissa, or the dynamic range?

Since essentially all mandelbrot shaders use FP, i would imagine the quality is largely dependent on dynamic range, so that FX12 would look far worse than FP16.

There's definitely alot of mantissa precision invovled. It's precision for texture coordinates that allow one to go very deep into the fractal. For this, dynamic range doesn't help at all. It may have an effect somewhere else -- I'm not too sure, though: I'm just delving into the Mandelbrot Set fractal image right now. . .