Irays most compute intensive surface options?
mike9
Posts: 69
in The Commons
I'm using a figure with a skin that isn't supposed to look like a human (which allows some freedom) and would like to shorten the rendering time. What surface settings have usually the biggest impact on rendering time with iray if they are active?
Also how big is the impact of mesh resolution->subD levels and mesh smoothing->smoothing iterations for render time?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2026 Daz Productions Inc. All Rights Reserved.
Comments
I'm pretty sure SSS is one of the biggest killers, particularly combined with refraction.
If the figure isn't in extreme close-up, and isn't supposed to be super gooey, you can probably shut off SSS and have a nice-looking image and save you a lot of time.
It's hard to guess what a "skin that isn't supposed to look like a human" means, but Iray dislikes "noisy" patterns, such as metallic flakes, as it extends the time needed to estimate convergence. Bump and displacement can also extend rendering, and as Will points out, this is dependent on the relative size of the figure to the camera. Far away shots render faster.
In any case, you'd get better responses if you posted an example of the skin shader you're intenting to use.
Actually, Tobor, you are bringing up an additional good point about far away shots and rendering faster.
What I was pointing out is that some detail (subd, SSS) is much less important if the character is further away, because you're just not going to notice the details. So you can, say, shut off SSS and speed up renders even more if a character isn't going to be right up to the camera.
While it isn't compute intensive, reducing texture sizes for objects that aren't close to the camera can also help reduce scene size (allowing more complex scenes to fit within the available VRAM).
If a character is only taking up 200 pixels of screen space, using 4k x 4k maps on them is just wasting memory....
SSS is one of the big computation hits. So is metal flake. So is refractive surfaces. Any of these can create a bit of a hit, computationally. All on the same surface = Major Slowdown. Emissive surfaces also add to the computational complexity, so each one adds to the computational load.
As Will says, "if a character isn't going to be right up to the camera" then it should be simplified, shading-wise. This will speed up renders considerably, and with a little effort, you can create 'lo-def' material presets for various characters/props/clothes/hair/etc for when they aren't taking up much of the camera frame.
Yes, you can save the presets, but you have to make them first. That 'with a little effort' is the part that seems to be left out of most of this type discussion. There isn't a 'click it and forget it' solution...it will take work and manual intervention to optimize. And each scene is different, so it needs to be done for each scene. And after a few dozen scenes, you'll have a decent sized library of commonly used items with hi-med-lo settings. Then it would be possible to use those to quickly modify not so commonly used items. One set of skin settings can be expanded to cover multiple skin textures in the mid-low ranges.
Thanks for the help so I got: SSS, refraction, noisy patterns like metallic flakes, bump and displacement, emissive surfaces.
Creating multiple material presets is a super idea I didn't think of as there will be scenes where the character is very close (face only) and some where he is in the background.
I haven't yet a final look as I decided to start again with the basis for 4.9. The figure is Genesis 2 with simple colored textures I create. The target is a look without much surface details somewhat similar to comic but still 3D and half realistic. But its also a general question as I ask myself every time when switching something like subD between 1 to 3 if it will impact the render time greatly for animations when doing it for most parts or if it will be just minimal.
My guess is decreasing the map size will also decrease the pre render duration. Anyone experience with this in iray and is it noticable or does daz keep the textures in the vram?
Emissive light is highly computational intensive, exponentially with additional lights.
I personally suggest that if you use this on something like a spiral bulb, or a light-bulb... turn smoothing on and use the lowest LOD for that item. The curves and additional projection surfaces seem to really stall rendering.
However, it is the most realistic form of light rendering, when compared to using the "Daz3D" lights, which are sort-of still a mix of 3DeLight and IRay. (Turning photometric on, also still renders the old rays PLUS photometrics, and the values are screwy per light type.)
One thing I noticed is that it will "complete", way before the scene is actually at 98% the target value. (Eg, full of noise, even after several long hours.) If emissive is your only source of light, or values are real low, while projecting through transparent surfaces. {Refraction, as stated above, I blame as the underlying culprit.}
Can Texture Atlas help with that? I thought it was still included with DS Pro?
Emisive/Emitters shaders is the most demanding iRay shader in my experience with iRay.
subD levels does not affect the render times, last night I tested a freebie, a simple bread prop with more than 1million of triangles (like the Godzilla movie 3D mesh)...Daz did render with ease in less than 1minute...and I was using SSS for the main surface.
Ok thanks. I did some testing and subD seems to impact for me mainly the startup time. If I set subD=0 it takes about 20 seconds until the the first preview is visble while with subD= 3 it takes around 1 minute 15 seconds. After that it seems to be similar fast/slow.
That sounds about right. subD impacts the work done when building the scene database, so the initial build time will take longer. Bump and displacement maps produce detail only during rendering, and require the renderer to calculate the alterations in surfaces on the fly. This is why these effect can consume more time during the actual render.
On emissive surfaces: if a scene must have these, there are ways to improve performance. First and foremost is to reduce the poly count for the geometry as much as possible. The more facets (even of the same plane) that emit light, the more Iray has to make additional light path calculations. You should also make sure the emissive surface doesn't have "competing" values -- for example, Glossy Layered Weight is generally not required for an emissive object. This is one of many nodes that are turned off when applying the Emissive shader. Some people take any old Iray shader, and then turn on the Emission Color channel in order to make the thing glow. That can result in really crappy performance.
Anything that causes Iray to make more light path calculations will slow down rendering, so things like refraction calculation, reflections, and caustics will require more render time. When rendering takes more than the default two hours to achieve a noise-free result, you can always up the render time stop-at value. As has been noted on the forums many times, Iray will stop rendering when the *first* of any of the following stop-at conditions are met: Max Samples, Max Time, and Rendering Converged Ratio.
The best solution to grainy images is to provide more light -- not necessary more light*s*, but more photons. Convergence is an estimate made by comparing subsequent pixel samples, as well as compariing nearby pixels. This establishes a uniformity that Iray uses to judge when a scene is fully cooked. By providing more light to the scene, there are more rays hitting surfaces, and the more rays, the faster convergence can be established. It's acceptable to "overlight" a scene and then compensate by using tone mapping. This is a common tech nique in real-world, too: A photographer might use over-lighting to achieve other ends, such as allowing for smaller apertures, and therefore, a greater depth of field.
Indirect and diffuse light are usual suspect in noisy renders. Iray has difficulty with these light conditions, so "helpers" such as well-designed HDR images, use of the Architectural filter, and other techniqes can help here. D|S still doesn't have one of Iray's most important indirect light helpers, sky portals. I was hoping this feature would be implemented in 4.9. Hopefully in 4.10...
Very interesting — portals are described in general terms (but not explained how to set them up) in the online docs for LuxRender. Unfortunately, they aren't implemented at all in the Luxus D|S plugin, although they are noted as something that might be added in the future. I'd really like to see what this feature can do in D|S. I'll join you in hoping for "maybe in the next version".
...so for example if using the emissive shader for all the lights in Jack Tomalin's Library, expect very long render times.
Short answer...yes.
One interesting thing to note is that including a physically huge set, even if it's low poly, increases render times. If the set covers five hundred by five hundred meters, it will take much longer to render than if the set covers twenty by twenty meters, even if the polycount is the same.
From my experience what will murder your render times is when two refelective surfaces in a scene interact and start bouncing light between themselves, something like two mirrors facing ecah other. Granted, this is excatly what is supposed to happen with PB renderers like Iray which is mimicking the real world physical behavior of light interaction with surfaces.
I did a simple test with a scene with a single-paned glass window, and a double-paned one. The double-paned window scene took more than 5 times longer to render. There wasn't much else in the scene to reflect apart from each other, and I would expect that time difference would really blow out in a more complex environment.
For scenes with multiple reflective surfaces you can control the ray bounces with Max Path Length. This is decided biased control in an otherwise non-biased renderer. Normally it's set to -1, which is "infinite." By dialing to 6 or 8, you restrict the number of bounces for each ray trace, and that will help some in rendering. Don't go much below this value, or else some surfaces will be negative impacted -- you'll quickly see it in eyes, since they use multiple geometries, and they all bounce light internally between the surfaces.
I've rendered a full Dubai Yas Marina imported from Sketchup and no problems with slower render times.
It seems that depending on any PC configuration it takes more o less time to rendering in different aspects of the scene, the only issue that covers all pc systems is using Caustics and Architectural, that options are longer in any PC configuration.