There doesn’t seem to be a specific forum where I could post this, since it transgresses rendering, phsyics, particles etc.
I’ve noticed that several settings (render samples, smoke resolution etc.) all default to binary values (32, 64, 128 etc.) whilst allowing any value to be used.
Is there any advantage to sticking with binary values (speed, quality etc.) or does it honestly not make any discernible difference?
Oh, weird. I just did a very basic test (sphere, plane and a single sunlamp), and rendered 3 times at each of 127, 128, and 129 samples.
129 took longer than 127 on all three renders. 128 took longer than both. Cycles renderer. Does cycles not like binary?
For splitting up work among threads or filling buffers it is beneficial to have the size be a power of 2 for dividing it into separate pieces, aligning data in memory etc. So for domain sizes etc it probably has some benefit. For stuff running from start to some arbitrary end like render samples it probably has no real nor thoretical effect whatsoever.
For 1 sample difference there are 99 other things that influece your timing. You should test it a lot of times to get any meaningful difference between them. The bitrate of mp3 playing on background or the web page you have open in your browser can probably influence your 1 sample time difference more than that one extra sample itself.
The three tests were all conducted under identical conditions (i.e. no background tasks that don’t have a constant CPU usage, but with regards to that, the renders were all done on a CUDA device anyway. No extra video stuff happening at all, and I didn’t even change the blender view, other than to change render slots. No materials, so no background computation going on.
On the test scene, the difference wasn’t significant (6.something seconds for each), but 127 and 129 came in at around 6.47 and 6.52, whilst 128 was (on the worst of the tests) 6.77. If we extrapolate that to a complex scene, the difference could be significant (I have one scene which is a 9 hour render, so this could be a 20 minute difference).
OK, ignore me. 10 more test renders at each resolution, and whilst 128 came in last more times, there were a couple of tests when it won. I’m willing to put it down to variation, in H/W availability, unless I decide to run comprehensive tests on something more substantial.