OTOY Forums

I have a question about emitter sampling rates.

Say you have a studio lights setup with three lights (please see attached image)...One light is red at 400W, the second is blue also at 400W, and the third is green but at only 10W. These three lights are equally spaced around a cube which sits in the center of the lights. Assume for argument sakes, that all three light colors behave equally and are not wavelength dependent.

Now, according to how I understand how Octane works, since the green light has such a low power in the scene, the cube surface that faces the green light will take longer to clear up noise than the other surfaces of the cube, since the green light is throwing out less "photons" than the other two lights. As I understand it, this is true if the Sampling rate of all three emitters are equal. So let's say that the rates for all three lights was "1". I end up with a grainy looking green side of the cube...Or at least it takes longer for it to clear up than the other sides. But if I raise the sampling rate of the green light to, say, "6", then theoretically, even at 40x less lamp wattage, the green side of the cube would clear up faster and not be so grainy in the end, correct? If this is so, is this faster rate of green cleanup at the expense of the rate that noise clears up for the other sides of the cube. In other words, since more effort is spent in Octane handling the green light, and the GPU only has limited resources, is that effort taken from the available resources given to the red and blue lights? I am assuming that all warps in the GPU are already full of tasks.

I havn't got any answer, but your question is very good... Can't wait to see some real answers

It seems i like answering these questions

Here is the answer to the best of my knowledge (Bear in mind i'm not an octane core dev).
It's a bit of a tricky one, tho' the quick answer is yes - it "steals" time from the other lights but it's not quite as bad as you may think.
It more like stealing attention / increasing the attention paid to that light.
Photon's are not sent from the light sources.
This is not how ray tracing generally works. Rays are actually sent out from the camera, bounce around the scene until the hit the environment or end because of max depth etc. The color of the pixel they represent in the final image is determined/contributed to by the results of it's travels through the scene.
So it actually works in reverse to how you may think.
Along the rays travels, attention will be paid to what light sources various materials are in direct line of sight to. The sampling rate changes how much attention each emitter gets in relation to the others. If a particular emitter is rarely paid attention to, then it will take more rays to get a good average sampling of what that light contributes to the scene. (and therefore be noise free).

Hope that makes sense...

Thanks
Chris.

no...

generally you need more samples if you have a mix of big and tiny emitters or complex emitters like ies that spread light from many points

or...

in case you have many lights and you want to give more importance to few of them

OK, then I guess the answer is that, yes, you can give more influence to the low power emitters, by raising its sampling rate, but that in terms of processing GPU instructions, do the high power emitters see a slight loss in performance? I imagine that the GPU warps have all been given an optimum spread of tasks and are therefore pretty maxed out. So something has to give in order for the low power emitter to be given extra attention.

treddie wrote: So something has to give in order for the low power emitter to be given extra attention.

it's not like giving out, it's more like balancing.. if You have big and small light..You probably get more noise from the small light source, thus resulting Your image almost clear..but not perfectly, 'cos around small emitter You'll have more niose..so if You cook image for longer engine keeps calculating light from the bigger object (but it's already good, noiceless). So You simply putting focus on different part in order to equalise the effect to the point where where noise from every emitter smooth out equally fast. You can take a look at this way that You take out somethign from big brother (ligh) for the sake of feeding small one, but big is already overweight =DDD so he doesn't feel anything bad.

That makes sense. Thanks for the clarification!