tricks for faster renders? (materials, lights)

B

bwrtz

0
Hey folks,

I'm sure this is a well traveled subject, but I'm looking for whatever dos and don'ts anyone has for materials and lights, in terms of what really bogs down render times the most. Someone mentioned sample size on skylights, which made a wildy significant difference in render times. Wondering if there are any other particularly pesky time-hogs that may be creeping around in the material settings or lighting settings. Or anywhere else, if you can think of it.

Thanks

Bill
 
When you say caustics, do you just mean the caustics tag, or should I also be concerned with the caustic settings of lights?
 
Especially refractive caustics eat a lot of processor power.

Also pbr-materials. Area lights are slower than the others. Complex reflections. DOF, a lot of samples.

Or to answer your question in short: Everything that makes your render look realistic.
 
I'm never using the caustics tag, that's a given. What is a "pbr" material? Would love to know which settings in the materials, lighting, and camera settings to avoid in order to minimize render time. What are the cheapest lights, what are the cheapest materials, etc. Good to note about area lights. Photon/caustic settings on lights? does that make a difference, etc.... Need all the tricks I can get. Already got the resolution and anti-aliasing down to the lowest I can tolerate, but I'm absolutely curious about materials (I am also doing testing on it). Thanks for the tips so far, I'll take whatever I can get.
 
pbrs are physically based materials (you'll find some experiments in the forum. One of the advantages is that you can download a lot of such materials for free, very often in very good quality and you can use them in almost any 3d app which gives a lot of compatibility).

Especially blurry reflections eat quiet a bit of processor time. And with simple materials the Cheetah renderer is faster than falcon (in Cheetah renderer specular can replace some of the reflections). And displacement instead of bump maps takes of course longer (as real geometry is produced here).

But to be honest, in my humble opinion you're on the wrong path here (as I already implied in my last sentence of the other post). If you avoid all the stuff that eats render time, you'll throw overboard all the good things. You will recreate images that were made in the 90s on cheap home pcs, not much too look at by today's standards. You know, cheap materials make for cheap looking images. I'm all for optimizing render times, don't get me wrong here, but for good results you have to pay with processor time. So you should look at that what you want to accomplish and use whatever tricks you find to get a good result.

So yes, a reduced resolution gives you faster renders (double the resolution gives at least four times the render time. But pictures in 640 * 480 are very seldom of any use nowadays).
 
Hasdrubal said:
pbrs are physically based materials (you'll find some experiments in the forum. One of the advantages is that you can download a lot of such materials for free, very often in very good quality and you can use them in almost any 3d app which gives a lot of compatibility).

Especially blurry reflections eat quiet a bit of processor time. And with simple materials the Cheetah renderer is faster than falcon (in Cheetah renderer specular can replace some of the reflections). And displacement instead of bump maps takes of course longer (as real geometry is produced here).

But to be honest, in my humble opinion you're on the wrong path here (as I already implied in my last sentence of the other post). If you avoid all the stuff that eats render time, you'll throw overboard all the good things. You will recreate images that were made in the 90s on cheap home pcs, not much too look at by today's standards. You know, cheap materials make for cheap looking images. I'm all for optimizing render times, don't get me wrong here, but for good results you have to pay with processor time. So you should look at that what you want to accomplish and use whatever tricks you find to get a good result.

So yes, a reduced resolution gives you faster renders (double the resolution gives at least four times the render time. But pictures in 640 * 480 are very seldom of any use nowadays).
Click to expand...

you may be underestimating how many good things are already in the sequence, and how devastatingly long the current render times are. compromise is an art that has been implemented by possibly every artist since the beginning of time, surely (especially if they want their work to actually exist within the actual bounds of time and space). but to make good compromises it's also good to know what all your options are, so you can be smart about which ones to use. having the render times exceed the duration of my lifetime, for example, is what i would consider the wrong path
 
I don't estimate at all, because I don't know what's there (and in all probability never will). I just see what's in this thread and go from there.

Like I said, I'm all for render optimization (I certainly spend lots of time on it). You know, for example using caustics or transparency and blurry reflection isn't a 'don't'; those are very important functions that sometimes are needed. So the optimization process includes also what you show (avoid everything made of glass, if possible, anything transparent, plastic with blurry reflections, and so on). But if you show it, you have to make it believable. If you go for a cartoon look, on the other side, the Cheetah renderer will probably be enough and help you to get faster render times.

And yes, compromise is a far better way than have render times that are going through the roof or bad materials only. But the ways you can cheat are fully dependent on the scene and include more than just this (like the geometry in itself (delete what's not seen), hdrs, texture trickery, the number of lights, and so on).
 
Hasdrubal said:
(delete what's not seen), hdrs, texture trickery, the number of lights, and so on).
Click to expand...
Please expand on this if you can, that's like my whole inquiry right there
Hasdrubal said:
If you go for a cartoon look,
Click to expand...
Not going for realism at all, (hint) so don't worry about that!

And i've already used plenty of transparencies, blurred shadows, etc, in certain places that i can't part with (must keep them, and i will). and
Hasdrubal said:
pictures in 640 * 480 are very seldom of any use nowadays).
Click to expand...
the rez isn't gonna go below 1280x720, and this is all going into final cut anyway so there's more to do after this. I just don't want you to feel like you need to protect me from doing something 'irresponsible' or 'unrealistic'.
that being said, if you possibly can i would love to hear more details about your last sentence in parentheses.

when you say 'texture trickery' i'd love some examples/specifics if you can. When you say 'delete what's not seen", i'd love some specifics on that (make invisible, or actually delete? 'not seen' meaning obscured, or behind camera? etc etc.). anything you can say about geometry affecting render times, because I was mostly under the impression (from my tests at least) that it was all about textures. Baking shadows, is that a thing? worth it? not? and so on. Whatever info you can give, i'd appreciate.

thanks again
 
> Delete what's not seen

The more geometry you have, the longer the render times (at least with complex models, because geometry fills the RAM of your computer. The more RAM you have the less of a problem it is). But it certainly helps, if you have a big scene, to make invisible (should already lessen the rendertime) or even delete, what's not in your shot. If you have a polygon-heavy scene it may be worth to really delete it because even what's not seen is somewhere in the memory. It's just not eating up so much of it. You would have to test (and in a movie a few seconds here and there already count) . Or f you have for example a figure with clothing, it's usually worth it to delete the body underneath (if you have a setting like this).

> Geometry

It could help to use low res models wherever possible (you could have several versions of the same thing). Again this gain is maybe in theory only.

> HDR

The complexer the HDR the longer the render times, especially with reflections. You can lower the resolution of the hdr and blur it (less detail, less computing time). Or even don't use HDR at all. It depends on what look you want and need to achieve.

> lights

Even with simple 'legacy' lights. The more of 'em you have, the scene gets more complex with each one which results in longer render time. And as said before: Area lights with soft shadows work slower than for example a spot, but you loose quality.

> cartoon look

Use Cheetah renderer. That's faster than falcon with simple materials (so you can use specular instead of 'physically (more or less) correct' real reflection. So for simple plastic, for example, you don't need a time consuming blurry reflection but a simple specular which is a lot faster and at least sometimes looking as good as the real thing.

> worry about ...

I don't worry. I sometimes only find it sad if somebody seems to go the wrong way and will loose sooner or later the joy of working with 3d. In your case, you seem to know what you want, but if the result would convince me is another matter (which isn't important in the least, by the way. You have to be happy with it).

>1280 x 720

That's quiet small by today's standards. For stills I use that as 'fast' test format (and the render itself will be much bigger).

> fps

You probably already lowered that, but I would go for the standard cinematic fps (PAL = 25 pics per second). You could even go down to 20 but some people see that already as flickering (there about lies the absolute minimum we can see a motion as fluid enough not to be disturbed by it. But better technology available made sure that younger people who were growing up with it will see a big difference between something with 20 fps and 30 or 60.

> texture trickery

I don't know what you try to achieve, so it is maybe not that easy to find simple pictures you can texture your objects with. But instead of modelling detail it sometimes is enough to have a flat texture on an object, just the pic (or painting). If you work with fake DOF (or do it in post), it's easy to use a texture consisting of a pic only in the diffuse. It doesn't even to be high quality if you'll blur it anyway later on. 'Fake depth of field' would mean that you use blurred textures from the beginning.

You have a shop with a big window in one shot? You can get away with a texture (you could even 'paint' the reflection for example with rendering one single pic of this window and then use that as texture).

It's faster to have a single plane with a pic and transparency (with an alpha map) than a complex model (but of course it can look very twodimensional if it's shown in the wrong angle or from to near).

In general, what's small and far away, you can get away with much less detail in geometry and material.

Instead of a complex background you could cut in some painted or photograpic background in post and render no background at all in Cheetah. Or, again, render a single pic of the background and use that.

You have some things that shadow your scene? Like some shutters or a tree? Use some texture map in the spot light (with alpha) and you have the same result in less time (or don't use that thing at all which would be faster).

Bump is faster than real geometry (but from certain angles you will see that's not real).


The part about using Cheetah renderer instead of falcon already should help to get much faster results with such simple materials. Specular is like a simple cheat while real reflection is trying to recreate the physical properties of a surface. For something like a cartoon this is more than enough.

But all in all it really is that simple: The lower the quality, the faster the render and vice versa.

And for each element you have in your shot you have to ask yourself: Is it worth it. Do I need that? And often the answer will be yes, because otherwise your 'world' looks sterile and dead.

For me careful planning is important, too. So I render things that will take longer while I sleep or do something different. In your case it would help if the computer is in use as much as somehow possible. If you know somebody else with a mac, it could help to ask, if you could render a few things in exchange for a license of Cheetah (ok, that one could get expensive, but if somebody is willing to help you with a few scene it could well be worth it)).
 
Image-based lighting is a huge win for rendering speed (Falcon or Cheetah) and actually helps with realism. In fact you can also do things like bake your lighting and background objects into a panorama, eliminate the stuff baked into the panorama, and render the foreground stuff faster (this is a common production trick). (There's a tutorial on this in my book.) This does not work for animations if the lighting changes over time, of course.

In general, Cheetah (the renderer) will be faster than Falcon, although for animatics, Falcon with low samples is faster and better.
 
Of the things mentioned above, the real killers are blurred reflections, blurred transparency, depth of field, complex lighting, caustics, and antialiasing mode (color can be WAY more expensive than edge).

Blurred reflections, blurred transparency, depth of field, and complex lighting all cause > linear increases in render time. (But all of these things are essentially free with Falcon.)

Caustics can be diabolical. Also, you're better off rendering reflective and refractive caustics in separate passes (in Cheetah). In Falcon, caustics are effectively free (you need to turn them on) but can cause scenes to render with serious artifacts.

Simpler materials are faster than complex materials, so using things like the metal and dielectric shaders when you can, can be a major time saver.

Things to worry about not so much:

High resolution textures — not much impact, unless you're using color anti-aliasing (to render the detail in those textures).

Image-based lighting (HDRI) is usually a time saver (as I stated above).
 
Please check this thread from last September:

https://www.cheetah3d.com/forum/index.php?threads/13042/

Increased C3D productivity can be measured in fewer minutes per render or more renders per hour. Tomas made an essential tool using the Mac’s built-in Automator and Javascript to sequentially render multiple files during downtime, like overnight.

Make speedy low resolution test renders with caustics off, modifiers disabled, etc., to tweak the lighting, camera, and composition, and then let Tomas’ tool make slower high resolution final renders while you take a break or sleep. It’s like a having a free 24 - 7 robot.
 
podperson said:
Image-based lighting (HDRI) is usually a time saver (as I stated above).
Click to expand...

Depending on the scene setup. With reflections ... not so much. Compared to a single spotlight, it's slower. Used as background, I agree; only you need quiet a bit of resolution, if the background should be sharp. Actually hdris are then a time saver when they are smaller and blurred (not so much to compute anymore).

There is a reason that something like SIBL was invented more than a decade ago (not available for Cheetah, but could actually easily be done), where you have a small, blurred hdr for lighting and high resolution ones as reflection maps (or even jpegs). The highest resolution is a jpeg for the background (but as only a fraction of it is shown in the actual render window, the available SIBL pictures have not a resolution big enough to work for todays high resolution renders (in this case, though, it could work). Stuff like this is faster, and that's why the workflow is incorporated in todays 3d apps, where you can control this stuff. It works only really well when the different pics work together. Something that Cheetah does not provide at the moment (it can be simulated with spheres).

That said, in my opinion HDRIs are more or less the only way to get decent lighting in Cheetah (often combined with other light sources). But how fast it is depends very much on the HDR used (I often use very high resolution where the hdr alone eats a giga of RAM).

so using things like the metal and dielectric shaders when you can, can be a major time saver.
Click to expand...

??? I thought they're even slower than the traditional materials to recreate glass and metal (but deliver far better quality). It would be faster to try not to use materials like metal and glass at all (or as few such materials as possible).

> High quality textures

They are not needed here. Some 2k is in this scenario more then enough (and for models they can be much smaller, very small if repeated).

Again, big pics can eat a lot of RAM.

All this stuff is very dependend of the scene. For example, images as backgrounds (and they don't have to be hdris when they don't contribute to the lighting) don't work with changed camera angles (or with the changed lighting).
 
Another way to speed up renders, which many of us have asked for, is distributed rendering over a farm, or cluster, of networked render machines. Cheetah 3D doesn't currently support this, but it would be useful.
 
Good stuff here! One of the best things to do is go to sleep while things render, even without Joel's suggestion.

And talk of distributed rendering brings back memories of a Mac II running three Radius Rocket boards.
Essentially three Macs running inside another Mac. Back then I was working with an old favorite, Bryce.
 
I agree that some type of distributed rendering would be nice.

C4d Octane 1920x1080 animation render I'm doing right now is averaging about 15 minutes per frame. So when it's done, it will have taken nearly three days. And this is using CUDA on a GPU with 8Gb VRAM. Some scenes just take a lot of time, no matter how much you try and optimize them.
Render Progress.JPG
 
You must log in or register to reply here.
Back
Top