[el.mustafa]

Dear All,

There is a basic issue with render settings that
I can't understand: How to get the best render times
by changing the polygon dimensions of your emitter objects.


Below is a Maya scene rendered in Maxwell.
Polygon count 2.4 millions.
Has one emitter polygon object and lit by a sky dome w/o sun




The default render at sampling level 12 completes at 2 minutes 38 seconds on 8-core machine




Keeping all settings the same but scaling the emitter object up by a factor of 2 in all axes
and increasing emitter power to compensate for the reduction in luminance the render
completes in 4 minutes 20 seconds.




At one point I was using emitter dimensions that made the scene render in 10 minutes
and thought, you know, it being Maxwell you'd expect that but accidentally realising
how million polygon count scenes can render so quickly just by adjusting emitter
dimensions makes me worry if my poor understanding of some Maxwell basics
has cost me a lot of rendering time unnecessarily in the past.

Regards
El

[Half Life]

At one point I was using emitter dimensions that made the scene render in 10 minutes
and thought, you know, it being Maxwell you'd expect that but accidentally realising
how million polygon count scenes can render so quickly just by adjusting emitter
dimensions makes me worry if my poor understanding of some Maxwell basics
has cost me a lot of rendering time unnecessarily in the past.

You said something I think can apply to nearly all of us there

It's not just size that matters -- it's also distance and power... but it's important not to overlook this does not just change the render times, it also changes the lighting results as well (which is partially how it is effecting the render times). Which means sometimes for the particular lighting effect you want there may be no way other than to have longer render time.

Best,
Jason.

[el.mustafa]

Hi Jason.
I wonder if Maxwell's reputation as an impractically slow
renderer was partly because of the user documentation.

e.g the massive reduction in render times when increasing
the global scale factor ought to be written in bold letters
for newcomers. Don't you think? So that users can clearly see render times
are comparable to biased renderers.

Add to that the extra time you spend adjusting Ambient Occlusion
(remember that one!) and what-have-you, you'll see that the extra
minutes spent on Maxwell do pay off in a way.


By the way, great tutorials.
Watched BSDF setting tutorials a week ago and for the first
time in 2 years I think I know what I'm doing. I even cut down
a lot on using sss and now able to approximate the look by paying attention
to Nd, K and the other standard properties.

Have a great weekend.
El

[Half Life]

Thank-you

I think there are certainly instances where Maxwells less-than-speedy reputation is well deserved

By optimizing your materials, geometry, and lighting you can make Maxwell perform like a speed demon -- conversely by being sloppy with your materials, geometry, and lighting you can waste much time. However some of the nicest effects Maxwell provides are also some of the costliest in terms of render time and so I believe there is always a balance between visual complexity/accuracy and render time that must be managed on a case-by-case basis.

One of things I found particularly time consuming in many other render engines was setting up scenes/materials -- often I found this takes almost as long as the Maxwell render time difference... and to be honest I'd rather rather let my computer work the long hours for me than spend all that time manually adjusting arcane settings . In that regard (once you understand the system) Maxwell can be a huge time saver.

I also believe that render time will become less and less of an issue as time goes by, through more powerful processing and improvement in Maxwell itself -- I see this as just part of the natural development cycle of technology. But having a stronger knowledge amongst the users will also help to minimize unnecessary render time as well.

Part of why I really wanted to do the Maxwell training series was I saw a real need for that type of instruction in the community -- In the beginning I had learned quite a bit from watching Mike Verta's videos on Maxwell Materials and reading many hours on this forum, but I felt the community could benefit from having access to a more expanded and organized format.

My intention is to keep working on as many Maxwell related projects as I can convince VTC to let me do -- I honestly think I could make this my full-time job and not run out of content because the possibilities with Maxwell are so vast and only getting more so as time goes by.

Best,
Jason.

[Mihai]

e.g the massive reduction in render times when increasing
the global scale factor

I haven't really seen this "rule" in renders, can you give more examples? It could be because the DOF increases when you double the scene size so the render then looks cleaner if there are less heavy blurred areas. In your example above, would be best to rule out any skydome/emitters combination and just check to see with the emitter alone in the scene what effect scaling has. Also you'll notice in the second render the emitter is actually a lot stronger. You can just switch the emitter material to lm/m² units, this way it will automatically change the amount of light emitted when changing size. Lastly it is sometimes misleading to compare two renders by SL, Maxwell could take longer to reach an overall SL but parts of your render can look cleaner than the render before run to the same SL. Better if you compare render time - let them render each for 3min for example.

EDIT: There could also be the illusion of exposure playing tricks in some cases - say you have a long tunnel with a weak emitter illuminating only part of it. At SL13 it looks ok, it illuminates part of the wall, the rest is black. Now you increase the emitter strength by 3x, render to SL12 and you see more noise. But it's because you are seeing more of the tunnel with the same exposure. If you were to increase the exposure in the first render to match the brightness of the second one you would see mostly the same amount of noise.

[Half Life]

He's not changing the size of the scene -- simply the size of the emitter, which being that it is a curved object around the subject should result in more complex lighting calculations.

Best,
Jason.

[Mihai]

Yes I know I was quoting his second post about increasing global scale would decrease render time.

[el.mustafa]

I haven't really seen this "rule" in renders, can you give more examples?

Hi there.
Whenever I use a material with any transmittance or subsurface scattering I always find it
much faster to achieve a clean result by increasing the global scale factor. The problem was
more pronounced in Maxwell v1 but anyhow I just did these quick renders in v2.5 to demonstrate.

Here is the scene with simple single-faced polygons acting as emitters and an hdri map:



Here is the render at 2min 15 seconds with scale factor at 1:



Here is the same scene at 2:15 with scale factor at 10 and compensatory increase in emitter power:



Depth of field issues aside, notice the anaesthetic tube in the area where it joins the
with its green filter compartment.

The global scale factor 1 image has poor translucency and more grain compared to
the global factor 10 following 2:15 of rendering.

I'd have to wait much longer to achieve the translucent look of the other picture.

Here is the material setting:



You could argue that the tube lies out of the DOF focus though.

I wonder if the problem lies with the emitters again. Whether I'm not reducing or
increasing power according to scale factor in a consistent way. That's quite possible
having read your tunnel/emitter example.

Ta
El

[el.mustafa]

Thanks once again Jason and please keep it up.

I'm particularly concerned with render times because I render
animations out of Maxwell. If the Ranch render farm start supporting Maya
then I'll probably just sign up for their unlimited rendering package
and turn up the sampling levels.

But for now I need to make use of any tip that reduces rendering time
even by a fraction. I do a lot of organic modelling so translucency and SSS are necessary
in all but the simplest scenes.

Regards
El

[Half Life]

Well in that case here a few you might not have come across:

1) Keep everything in focus -- DOF is slower to render

2) Use polygonal aperture for your camera it supposedly renders faster and with no DOF you wouldn't see bokeh anyway.

3) Avoid/disable motion blur.

4) Set your "Min. Time for saving to disk" to 60 min in Maxwell.

5) Use the -nomxi flag if you don't need an MXI file.

6) Use the -nogui flag if you don't need to edit the file while it renders.

Each of those may only shave off a little bit of time, but over the course of many frames it may add up.

Best,
Jason.

[Mihai]

El, you're reaching a wrong conclusion I'm afraid. If you make your objects ten times bigger ofcourse this will change entirely the look of any tinted glass or SSS material. Although the material stays the same, light now passes through a much thicker object so ofcourse you'll see more attenuation. What you should aim for are material settings that look good for the intented size of the object. In the case of that green filter just decrease the attenuation distance which is much too high (8cm) if you want a heavily tinted plastic for such a small object. Notice 10x size one also has noise, the smaller one seems to have less because it's clearer - you also see noise through the clear parts which are more heavily tinted in the 10x version.

Also your ND for that material is too high - should be around 1.4 - 1.5 for a common plastic.
I don't know what's going on with the teeth either, they are almost like emitters. Probably you have too high refl0 brightness on them.

(there are good SSS presets in the mat wizard you can use as a starting point until you get more familiar with how it works)

I think it would be good for you to learn more first of all about the material system, try to model everything as in the real world (btw it would help your models a lot if you add some small fillets, like on that green filter, to make them catch the light in a realistic way), set your emitters to real world values. Once you have these realistic materials, models, lighting, camera settings - you'll get the best and most predicable results from Maxwell. Then it's a matter of tweaking camera settings if you want less blur from a narrow DOF....ofcourse you can cheat a little bit if you're rendering a really small object and want to avoid blurry parts. Set first the fstop really high like 22, if that's not enough scale the scene 2x. But scaling it 10x like in this example, that head is around 3 meters big.....your lighting will look all wrong.

[el.mustafa]

Thanks Mihai

That makes things much clearer.
One is still tied to the non-physical rendering mindset.
Sacrificing accuracy for aesthetics when needed (and often achieving neither because you don't want to wait for Mr Renderer to sashay his way through the best part of the day!).

Mind you, the teeth look awkward because there is a second emitter inside the mouth.

And don't worry about the Nd I forgot to attach the base BSDF which is:


Thanks.

[el.mustafa]

And thanks to Jason once again for the top tips