# Same rendering gives different noise amount

I usually render my single frame projects with the SheepIt-Renderfarm as animation of low-sample frames. But I experience great differences in the noise amount between some of the frames. For example, seed 5 (top) and seed 41 (bottom):

As you can see, one seed is way noisier than the other. There are several seed with less noisier result and the quality is always the same, like if some of my frames had been rendered with twice the samples.

Why are some seed noisier than others? How do I find out those? Is this even seed related?

I also tried the "split frame" function of the farm where each frame is rendered in 4 different tiles. [Imagine a rendering like the seed 41 one, but with seed 5 in the bottom left half of the picture - I'll add it later] There is the same effect, but on a single frame. So: same .blend file, same frame, same seed, different result. Why?

• That is really odd. It's definitely NOT seed related, that's for sure. I've rendered plenty of animations on my own machine where I animated the seed value (different seed for every frame) and all of them had the exact same amount of noise. I'd be more inclined to believe that this is an artifact or bug on the render farm, than that you've uncovered some magical seed values ;-) – Matt May 10 '16 at 17:38
• Then it must be hardware related. But how could that happen if the same blend file runs the same algorithms on a different hardware, and why should this result in different noise amount? – piegames May 13 '16 at 12:35
• Sheep-it uses average users machines, each is setup as the user wants it, maybe some users have a setting in their startup.blend that doesn't get overridden by the render task. I'm thinking square samples as one setting that would give this variation. – sambler May 13 '16 at 12:38
• Sheep-it uses an own Blender installation that is new for every session. All clients use the same Blender version for a rendering project. – piegames May 13 '16 at 12:40
• @piegames that's right, the "average user" can't even access the copy of Blender that Sheep-It is using. You're right, even different hardware shouldn't matter. It's REALLY strange. It'd be really nice if we could get a Sheep-It developer in on this conversation... Would that be out of scope? – Matt May 13 '16 at 13:02

I'm not sure i have an explanation for that.

The faulty frames were rendered by two different users (5 for one user, 1 for the other), on different hardware (4 cpu, 2gpu). Only "windows" was the same but i don't know which version.

The Blender version used by SheepIt is actually the one from Blender Foundation, I just remove the blenderplayer.

Here are the binaries to verify, if that helps https://www.sheepit-renderfarm.com/media/binaries/

I'm also not modifying any seed value. Here is the python script I uss on a CPU render:

import bpy
bpy.context.user_preferences.system.compute_device_type = "NONE";
bpy.context.scene.cycles.device = "CPU";
bpy.context.user_preferences.system.compute_device =  "CPU";
bpy.context.scene.render.tile_x = 32;
bpy.context.scene.render.tile_y = 32