SunFlow Quality Test (diffuse shaders)

Investigating different Anti-Alias settings in SunFlow. Be sure to compare the output at original resolution.

 

I previously tried out the Ward-shaders in SunFlow - which turned out to be slow - taking several hours for a single image.

 

These images are rendered using the 'diffuse' shader, and various AA settings. As can be seen they are much faster.

 

I also tried out a version of SunFlow compiled using the Java Native Jet compiler - something which in principle could make Java programs run faster, since they are compiled into native code, instead of being interpreted/JIT-compiled. The version I used was compiled by PolyQuark and can be downloaded here: www.polyquark.com/opensource/.

 

As can be seen from the table, it actually performed worse than the standard version (~40% slower). The image output was exactly the same as for the standard version (as expected). Notice that the native compiled version was 0.07.3, whereas the standard version of SunFlow is 0.07.2. I don't know why it turned out slower - I just tried it out without changing anything.

 

Rendering time:

 

AA Time Relative speed

(-2,0) 0:34 1

(0,0) 0:47 1.38

(0,1) 1:54 3.35

(1,1) 2:48 4.94

(1,2) 6:44 11.88

(1,2) 9:25 16.61 (NATIVE)

(2,2) 11:19 19.97

 

Time in minutes. Rendering was done on a Dell Vostro Laptop (Intel Core 2 Duo T8100 at 2.10 GHz).

 

Some scene details:

Pinhole camera, two meshlights, a single diffuse shader for all objects, 498x450 pixel output, and otherwise default settings: "accel bih, filter mitchell, bucket 32 row"

 

My conclusion: AA=(1,2) is very pretty - no need to go higher. For quick previews (-2,0) seems like a good choice, rendering at 1/12 of the time.

Done