Iray: what factors directly affect iteration speed?
evilded777
Posts: 2,444
Seems like a simple, obvious question... but its not.
Lets use two common, pretty sophisticated examples: Stonemason's Contemporary Living and A Bright Loft. Both pretty complex scenes, lots of glass and reflections, etc. But Stonemason's set will render in easily 1/4 (arbitrary number) the time as A Bright Loft. Why?
This is not intended to compare the props or the artists who made them in any way, nor to laud or disparage either. This is for my own edification, so I can USE the products I own, hopefully, as I want.
Comparison: A Bright Loft after 3 minutes, Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 00296 iterations after 185.781s. and less than 1% done
Contemporary Living after 3 minutes Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 00511 iterations after 181.841s.
So not quite twice as fast, but A Bright Loft is at less than 1% converged at this point, where Contemporary Living is >6%.
All render settings are exactly the same, not extra lights, only the default HDRI image.
What elements lead to one rendering that much faster than the other?
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
Rendering with only the default HDRI and an interior set? First of all, you really should be rendering with the Architectural node enabled in your render settings. As for the time it takes for both of these sets to render with the same setup, the answer is probably the size of the windows allowing light into the rooms. Stonemason's set has very large windows on two opposite sides of the set. A Bright Loft does not have nearly as much windowspace to allow light in, therefore it takes many more iterations to render.
Points noted, Chuck.
Subtract them from the equation. I used the Architectural sampler. Available light is NOT the issue. I have run numerous tests, adding light to certain sets via primitive/emissives and actual lights. They STILL render slow. (The Locker Room is a good example; and I don't want it to seem like I am picking on Bluebird, I own most of their products and adore them... I just can't get them to render with any speed in Iray).
I did a render of Stonemason's Utopia Deck C using pretty much ONLY the modeled lights that are part of the prop and it rendered brilliantly.
Maybe I am way off base here, but it really does seem to me that there is something intrinsically different that makes some sets take longer too render.
To step away from the initial two vendors, I just did a test using DZFire's Armory prop. No global lighting, it is completely enclosed. Using only the available emissive surfaces it plunges along and after 3 minutes it is 14% converged and has passed 1600 iterations with little available light to move it along. Granted it doesn't continue to converge at that rate as the render progresses, but the iteration speed is still there.
So... perhaps the question needs to be rephrased: what directly affects iteration speed?
Well now... perhaps I have misinterpreted previous tests.
Because available light does seem to be a larger factor that previously considered.
Performed the same test on The Locker Room as The Armory, using only available emissive surfaces and it rendered much faster and much cleaner (it has BIGGER Lights and less area too light than The Armory).
So there is an obvious flaw in some of my thinking, or experimentation process.
Side note: I thought adjusting Exposure Value was supposed to adjust more factors than simply shutter speed? Because for me, right now, that's all its doing.
There are many, many reasons why one set may render faster than another...... Overall poly-count, number of trans-mapped surfaces, size of the textures used, number of texture maps used per material (for instance one texture might only have a diffuse and bump map, while other texture setups may include specular, opacity, ambient, reflection, displacement, translucency, and subsurface scattering maps). Shader settings are also going to be a factor. The complexity of the scene is a major factor... light is going to bounce around an empty room much faster than a furnished room, and more complex setups are going to take longer to render, even with low poly objects.
If you are using only the emissive surfaces in a room to light a scene, try this... Leave the tone-mapper at default settings and set the dome mode to "Scene Only". When you set up you emissive surfaces, right under the "Lumens" setting there is a setting called "Luminance Units". By default this is set to "cd/m^2"..... switch it to "kcd?m^2"... the "K" setting multiplies the Lumens setting by a thousand. Don't be afraid to really crank up your meshlights !
I have this vague idea that there is a way to get information on a scene's polycount, but I do not know what that is.
I should think that gauging the memory used when rendering ought to give some sort of equivalence for disparate props, no?
Chuck, I prefer to use tone mapping and "realistic" lighting values than to simply throw light at the problem. Its a preference.
If we are talking about two separate sets that each only have one texture option, checking the size of the installer files might give you some clue as to which one is going to render faster... but the size of the scene is not going to make nearly as much difference as the complexity of how everything is arranged within it and the materials and shaders used.
I can understand that very well.... but from a cinematographer's point of view, that really doesn't work for artistic images. When you see beautifully lit movies like the "Lord of the Rings" trilogy, you have to note that the brilliant sunlight streaming in through the window is usually not natural sunlight... it's artificial, on a soundstage, and it's brightness is whatever it needs to be for the scene. There's always a certain beauty to something that is done as realistic as possible, there's no denying that. I like stuff to look as real as it can too, but I really enjoy "painting" the scene with light. Everyone has their preference ;)
Here's something interesting, from the Docs
SS RGB Unit Conversion - Allows the setting of the units, described in more detail below.
The properties in the Sun-Sky group are designed to enable physically plausible daylight simulations and very accurate renderings of daylight scenarios. The value of these properties are measured in true photometric units, but the output can be converted to something else with this property.
If set to 1, both the values for the sun and the value for the sky can be considered as photometric illuminance values in lux.
Since the intensity of the sun outside the atmosphere is calibrated as a 5900º Kelvin, blackbody radiator providing an illuminance of 127500 lux. This is very bright when seen compared to a more “classical” rendering where light intensities generally range from 0 to 1.
The value of this property is applied as a multiplier and could be set to a value below 1.0 (e.g., 0.001) to convert the raw lux value to something more “manageable.”
For convenience, the special value of 0 is internally set so that 80000 lux (approximately the amount of light on a sunny day) and equals the classical light level of 1.0.
An interesting alternative is to set this value to 0.318 so the final rendered pixels in the image are true photometric luminance values in candela per square meter 2).
These true luminance values then fit perfectly as input to the photographic Tone Mapper with its cm^2 Factor set at 1.0.
If you are blending interior and exterior lights, try the .318 setting. Wow, what a speed difference.
Open the Scene Info pane. With nothing selected, it will give you the poly count for the entire scene, or you can select individual items for their poly counts.
mac
Materials are very relevant to render time. I was trying a jade statue on a wooden platform and the platform was visually perfect after about 300 samples while the jade was still grainy at 1300+ samples.
I've wondered about that SS RGB unit conversion too. I've noticed that when I'm trying to use sun settings from an sIBL set that it looks better with it at 0.318. It seems like there's a disconnect between the defaults, either it should just be 0.318 by default, or the tonemapping settings should be changed to work best with the unit conversion at 1.
I would think that the unit conversion should not work with an HDRI as it is specifaclly for SS or Sun & Sky...I have not tried it with an HDRI.
Does anyone now what the Architectural Sampler actually does?
Has anyone tried simulating sunlight/daylight with an emissive? I took a stab at it, specifically for what Chuck mentioned above about sunlight through a window. Nice control, not sure about iteration speed improvement.
Surface or material settings definetly can have an affect on the amount of samples required, as latego mentioned. Do they also affect the speed of sampling?
The architectural and caustic samplers are explained here: http://blog.irayrender.com/post/51722647664/the-architectural-and-caustic-samplers