IRAY: Speed x Size
Scavenger
Posts: 2,664
In 3Delight, a rule of thumb is if you double the size of your picture, you quadruple the time, due to 4x more pixels to be rendered....
Does that apply to IRAY too? Since it doesn't do pixel by pixel, is it different?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
I remember someone posting that their iRay renders were actually faster when they increased the resolution, though I think this is highly dependent on the scene, and in many cases will be slower, but probably not 4x as slow like 3DL.
Faster...sort of.
You can double the size and render to a lower convergence, then when you shrink much of the 'noise' goes away...at least that's the theory. But with Luxrender I never had much savings with that technique.
Greetings,
"Manual Oversampling" this article calls it.
I've been doing that with most of my renders, when I don't want to render all night (which I am very often comfortable with).
-- Morgan
Huh..that seems to work..haven't done a speed test, but killed a tripple size render at 33% and it came down looking good.
I've been playing with "manual oversampling" but I'm wondering if anyone has done speed tests comparing (size/convergence) to (2xSize/.5xConvergence)
For sure, I'm seeing it seems to take longer for a bigger version to begin actually drawing, but my renders fall in the multiple hour span, which makes true testing rather...difficult.
A smart despeckle filter might be better and faster than these methods. The worst noise appears in darker areas and very bright areas, so if the filter could concentrate on these ranges, it could then apply a despeckle process which would identify dark pixels immediately surrounded by lighter ones, and then apply the average brightness of its neighbors. This would only work if you've let the image reach, say, 50% or so convergence. Below that there may be too much noise and you could lose sharpness in the details. Most images reach 50% well before 50 percent of the render time. It's the last 50% that takes the longest.
I'm just as curious as everyone else on 'manual oversampling'. That linked piece talks about specific times which is generally meaningless because he probably has a superfast rig and has been using iray a long enough time to know a certain size image with the specific refraction stuff he put in there would take about a certain amount of time. But we don't have that experience and experimenting can take a lot of time especially for those of us with CPU-only on older rigs.
So until we have more experience, setting times for a render is rather fruitless. So the other option is (after reducing samples in render settings) to also reduce max iterations?
But as for 3delight 'double the size, quadruple the time'. I don't think that's really true. Not in my experience anyway. It's more like a tad more than triple. That's a bit subjective, though, because only once or twice have I ever bothered to write the times down to compare.
Quite a while ago, I ran some tests on that in 3DL...and no, it isn't exactly as you'd expect it to be. Double does not necessarily equate 4x the time. Alot of that depends on what you have in the scene and what shaders you are using. Simple, not complicated scenes without GI and SSS, yes, that's pretty much true. There's not a lot of 'prerender' calculations going on that are going to add to the time. But once you start adding in things like SSS, full GI and volumetrics...things that rely on a lot of precalculations that aren't tied to image size, but rather on the geometry, then it can really save time.
With iRay, you can double the size and it can be quite fast. But it'll be much much slower if there are more dark areas. If it's a well lit scene, the increased time won't be that much. I tried various attempts at the manual downsampling "trick". It only works for low convergence. IOW, it only helps if the image still has some graininess to it. If you want a really high quality render, you won't gain anything by rendering at twice the resolution.
Depending on the scene "convergence" at 95% may not be substantially better than 70% or 80%. Haven't been able to notice a correlation. However, I have not seen improved quality by changing from quality=1 to quality=5.
FWIW, I think a number of people still think of Iray as some 3delight clone rather than a distinctly different engine. I also suspect that it is not so similar to Luxrender or Reality either. I only have personal experience with 3Delight and Iray. Point being Iray needs different techniques and methods than 3Delight.
Can confirm this technique. I normally render 1920x1080 until I read this technique a few weeks ago. Tried it out and it's surprisingly true. I find it doesn't really help below ~10% convergance, though.
As far as convergences go..I've done renders where in order to really see differences between 1% and 10% Is to layer them in photoshop, zoom in 400% and turn a layer on and off and then you see pixels move.
Hmmm, mine are pretty stark between 1 - 10 and then after that it's small until it finishes