GI Render Settings


Right fellers, Ernest B was thinking about this thread i think, but i thought i’d post it anyway.

GI Render Settings. What about them? i though we’d have an informative learning thread which will hopefully give us all a better insight and use out of our renderer.

Im a proffessional architectural visualiser and switched to C4D as my main render weapon from max about 3 years or so ago, and have always been lucky to have dual xeons provided for me as my render processors, so maybe i’ve picked up more bad habits about the understanding of what i’m actually doing than i should.

out of interest, here’s a quick example of a wip of the typical kind of thing i work on. this’ll give you an idea of the general complexity of a typical stand-alone model (sorry i’m not at liberty to provide test models) -

This particular scene has about 50 internal lights (w/shads) and a few external. Accuracy set to 100%, stoches = 100, min=100 and max=100, with a dd of 1 and a prepass size of 1/1. it rendered in 4 hours at a res of 2500 pixels wide on a dual xeon. (not too bad imo)

I’ve always gone for as high an accuracy setting as i can, with lower samples. i’ve always found this to be very visually pleasing as possible (in my case anyway)

I’ll typically use an accuracy set anywhere around the region of 95% +, with a stoch sample rate at around 100. my min samples usually range around the 30-50 mark, and the max samples are to suit the model, but again, usually around the 100-200 mark.

(I ALWAYS set a diffuse depth to 1. But this is personal choice - i prefer to keep a nice render speed and can successfully use infill omnis to compensate for lack of bounces, and i’ll use odd AA settings to for optomisation. But lets not argue about this, as it almost warrents it’s own thread.)

I’ve always found the combination of accuracy, min/max and stoches easy to use and set up, but not always understandable. particulaly the min settings.

to me, min settings are relevant for larger sparce un-interrupted areas of a mesh and max settings for intersections and complicated bits. Accuracy and general stoch samples for more of the realism side of things.
i.e. i’ll set a general min/max settings to get a non-blotchy image, completely indipendant of what i might set the accuracy and stoch samples.

but of course, they all work together too. but how can adding more min/max samples up the realism? all they seem to do is iron out the fiddly splotches. it’s the accuracy and stoch samples that determin the realism i thought.

why not use an accuracy over 90% as some peeps recently in here suggest you should never do? whats the point of the option to do it then? i use it to good effect :shrug:

i’ve been testing yet again with a lower accuracy and higher min/max/general settings and am getting nice results, but not as real as when i slap the accuracy up. also, the render time obviously shoots up the more samples i use.

so what do you think. what would you suggest for that model above there, with in view for realism of lighting.

the more understanding we can get the better.

cheers for your time. hope i havent drivelled on too much :stuck_out_tongue:


Well on my side, during the various stages of testrenders, i gradually lower the min and max values up to the point where artifacts/splotches get visible - then up it again a little and have my final rendersetting.
Also note a value of 100 for min seems alot. I rarely end up with a min value around 30 - mostly its under that.

About the same procedure with stoch samples.

To Speed up Testrenders i reduce accuracy and the preview size to rediculously low numbers.
I usually end up with an accuracy between 80-90% for the final render as anything above that slows down the render tremendously.

I don’t use many fill lights to enhance an image since it takes a good deal of work to match these against the overall GI. If they dont match the GI perfectly the whole image suffers and that is not tolerable.

Also the nescessary tweaks to the materials to get good reflections/surface properties play a major role for the time a render takes to finish so if i can get away without such calculation heavy materials it is very fortunate. However since i mostly do closeup shots these materials are a nescessity and add alot to rendertimes.

With the new release (better area lights/shadows/blurry reflections) i believe we will have lots of new options to discuss. In the meantime this is my take on GI rendering.



Strat, thanks for taking the time to keep this discussion going and very nice work. A couple of questions:

  1. Is your setting accuracy to 95-100% a part of a tried and true production formula or is there a perceptible quality improvement that others can see if pointed out? If it is perceptible, could you use this rendering and provide an enlarged and rendered small detail with accuracy set at 70%, 85% and 100% (assuming the other render settings would have to be adjusted accordingly)? Me, trying to be scientific. lol

  2. What is your opinion about MJV’s accuracy and c tag head room comment:“You should always have it set to 70% in order to leave yourself some head room for changing the value in the compositing tags.”?

(BTW, thanks for the valuable help you’ve given me in the past.)

Edited 2.


I just noticed you used a diffuse depth of 1

Next you likely put over 50 lights behind your windows to illuminate the rooms right?

And an accuracy of 100?

How about a dd of 3, no extra lights, adjusted min/max values and a reduced accuracy of 85% - would you care to try out this configuration and let us see the outcome (time/quality)?

Of course for such a discussion it would be beneficial for all participants to have access to the test scene but since you can’t hand that out i’d like to see some of the suggestions that might come up incorporated in testrenders from you - if you have the time of course.


I’m thinking this would make a nice mini challenge. Someone would have to supply a scene though. I guess everyone would submit their best settings to optimize speed vs. quality. The downside would be that in order to judge someone would have to test all the settings on one machine. Or maybe there would be some way to standardize render times by figuring in the individual entrants’ CB scores?


That sounds like a good idea. A CB score as mulitiplier.

There is always the scene I made for the glass tests. Not as exciting an example of arch-vis as STRATs sample, but the model is available and has an exterior and interior component. Just a thought.


I guess the CB would a divisor, but I think it would work okay. :slight_smile: It wouldn’t be 100% accurate because CB doesn’t cover every situation, but I think it would be good enough for such an informal contest.

I think I still have that scene you posted–will have to check. I’d suggest using only two tex in the scene: default and glass. Default is good for showing artifacts and glass is good for reflection and interior lighting issues. Of course everyone should use the same glass material.


That scene was one of the first things I made in Cinema, so if it would need any work (joining surfaces, etc) I would not be insulted. If you or anyone wants to use it that’s fine, since testing is why I created/posted it. It’s still a pretty boring little street, unfortunately.


STRAT and all, I’d be wiling to model a simplified version of your scene
and, or you could provide one with just some primitives stacked up and
selection sets to represent the glass.

I feel that having basic lighting and camera angle matching the test scene
to be critical as well.

As per other comments: we’re just adjusting settings on our own machines
to test time and variations of render quality.

As a side note: When getting mildew splotches in an interior scene I found that
if I cancelled vol lights, the splotches went away. However, additional fills had
to be added to make up the difference.


damn! a busy day. havent had much time to check this thread.

but, an easy day planned tomorrow, so i’ll do some test renders and details as requested and post up results.

cheers dudes, keep it comming…


dudes, i spent all day yesterday rendering and testing on a live job.

and yes, a lower accuracy and higher samples is a nice combination. faster too. thanks for the advice fellers.

old and usual settings -

strength - 100%
accuracy - 100%
prepass - 1/1
diff depth - 1
stotches - 100
min - 30
max - 100

render time over 4 hours.

new settings -

i’m generally getting lovely realism, slotchyless too, with settings like -

strength - 100% (i’ll always use a 100% value)
Accuracy - 80%
prepass - 1/1 (depending on scene type)
diff depth - 1 (again, i always use 1 for fast speeds. i’ve a good infill light backup to compensate)
Stotch samps - 1000
min - 300
max - 100

the image at the front of the post now renders cleaner, just as realistic, but in 1 hour instead of 4.

here’s the new pic, a touch of post photoshop, and a lense flair. yes yes i know, but i thought it worked nice in this particular instance -


ok ok, but the kind of lens flare that you choose don’t match the kind of camera that you have. The visible ring it must be more trasparent but more visible the white circle on the centere where the window is. Please consider this comment als a personal feeling.

Very interesting thread and I always was suprise about your settings for the renderings… excellent results.



Thats a really big step down rendertime wise Strat. The quality looks great. I like the way you’ve merged the photo in too. It looks almost seamless.

Just wondering what settings everyone uses for stochastic mode radiosity?


Gee, I don’t know - using GI at all is soooo five minutes ago limp wrist


Seriously though, I like the challenge idea, and using CB score as divisor should be accurate enough me thinks :thumbsup:



whats CB score mean?


The rendering speed mark you get when running CineBench.



Nice job Strat … I would still be curious to see how dropping the 50 internal lights and increasing the number of bounces effects the quality and render times. Just in case you had a few free hours … lol.


Somebody not had his coffee yet?

CineBench, as a way of equalizing time results on render tests to a set resolution.

I’ve not run CineBench since before I even owned Cinema, but its a great idea.


that render hasn’t actually got the 50 internals switched on :thumbsup:

and nope, at that high a realism setting and resolution there’s no way i’m adding an extra bounce. you can garantee the render time will massively increase. the difference in render times between 1 and 2 bounces really is significant.

besides, i’m more than happy with adding infill lighting to compensate for extra dd. just my choice :slight_smile:


i like the new render better too - it is good to see that c4d DOES have the tools necessary to do a nice render at a good speed - it is just the settings have to be adjusted to get there

a 75% render time reduction is certainly something to justify a bit of time doing render tests… update us as you see if these “revised” settings benefit your other projects.

good to know, and good that you posted so others can see “it can be done” : ) (not always maxon to blame)