Adding oomph to your rendering time

[banditcameraman]

I happened upon this technique by accident and after having tested it under different circumstances, thought I’d share it. It a variation on CPU-Accelerated so if the aliasing bothers you don’t use it. If your scene has areas which are very dark (pitch black) or is a mixture of very dark and very light, this may be even more effective than CPU-Accelerated with Extra Boost but you are going to need about an hour and a half, depending on the scene.

I won’t be embedding examples here because of size limitations but linking to them for you to see and make up your own mind. When an NSFW warning is needed, I will flag it.


The technique:

Is basically simple: (huge props to sigstan ) This means using "export only" and tweaking the code. It’s not as hard as you might think and is ideal for that because it’s free; with the file accessed through Notepad ++, which not only has the advantage of being able to keep the tabs of different files open, it will also keep the selected part of a file highlighted so you can just copy-paste from one file to the other

Example 1:

A very heavy scene: G2F, HiveWire Horse, HiveWire BigCats have all been subdivided to 3 in D|S and with the exception of the leaves and small rocks in the scene, everything else has had displacement maps and subdivided into 2, on the average, in Reality. It takes LuxRender about 10 minutes to load the file.

Rendering time and resolution: 10 hours to reach 13 k s/p, with network rendering and overexposure to 6400 while rendering.



Example 2 (NSFW):

One G2F figure and SAV Mitchel Hair, each subdivided to 3. Light comes from Fuzzy70’s SingleIBL and Reality Mesh light provides a backlight.

Rendering time and resolution: 1.5 hours to 6-ish k s/p on one day, another 1.5 hours on another day for another 6-ish k s/p and an additional hour on another day for a total of 13k s/p with network rendering and overexposure to 6400 while rendering.



Example 3 (very, very really uber NSFW):

4 G2F figures with Mitchel Hair, each subdivided to 3 in D|S, light from Fuzzy70’s SingleIBL and a mirror floor.

Rendering time and resolution: 8 hours to reach 33K s/p , networking rendering with no overexposure



Near as I can make out, the combination of CPU-Acceleration and Power Light Strategy should be 3 times faster than CPU-Acceleration alone. On the average the rate is 5k s/p every 1.5 hours. YMMV

As always, this should be considered a starting point

[sigstan]

Very interesting observations!

Not sure why the Power light strategy is faster in these cases, but it could have something to do with the light calculation being much more simple, if the raw power ratios are used.

Thanks for sharing this.

[banditcameraman]

You're most welcome and thank you very much and huge props for sharing the technique, sigstan

Because of the heavy rains our internet connection has been down since 4.30 pm yesterday (it's now 11.30 am the day after so that sucks ) and am replying from an iPad in an Internet cafe

Re-did the example 1 because some minor tweaks were needed but instead of waiting till the node was tessallating before adding the next one, I added each slave as soon as the previous one was accepted. This resulted in better rendering time and it reached 13 k s/p in 7.5 hours instead of 10

And the light setups are fairly simple with either Fuzzy70's singleIBL or DzFire's Torch construction kit and a mesh. I have another scene with more complex lighting and the power light strategy doesn't seem to be as effective but having said that, I need to let it run longer: with the default light strategy there is no improvement from 11 k s/p to 18 k s/p, which should not be the case. Will try another light strategy when the Internet connection is restored

[banditcameraman]

I've had a chance to test the Power Light strategy more. Granted network rendering will give you insane rates (more on that later) but even using just one will give you 600-1K s/p, depending on the size of the render/light set up. The only way to know for sure is to test it, whether on a new render or to re-render an old project and see if the performance increase is noticeable.

Example 4, safe for anything but I just didn't feel like resizing:

2 HiveWire Big Cats, light is Fuzzy70's singleIBL.

Rendering time and resolution: No overexposure, network rendering- 46 minutes to reach 4.6K s/p

[banditcameraman]

I was going to bed but after a PMing with nathanomir, thought this might prove useful

Here's the render scaled down to fit the forums:



.

Yes, I must have been insane to turn the walls and floors into mirrors That was done out of necessity as I couldn't get the results I wanted when the walls were set to its original glass. The set is .

Anyhoo, after setting the lights and camera, I cranked the ISO up to 6400 and let it cook. I knew I was on the right track the first day when it started to look good. Previously when I tried it without the Power Light strategy it still looked grainy at 10K, even with nodes. With the Power Light Strategy in the same amount of time I hit 20 K s/p. Still grainy, but developed enough for me to see where it was going.

Rendering time and resolution? 5 days to hit 105 K s/p. Yes: 105,000 samples per pixel

And that's not what you see above: the high key version looks good at 105 K s/p, because the hall and wall lights are set to a gain of 9000. I was going for a low key look and at 105 K, I'll probably need another 50 K. But the high key looks decent and has a sterile environment vibe.

Still going for the low key, though

So yes, if you use Power Light strategy, have nodes or a rendering box (anything to network render to that will amp up the master's processing power), and crank up the ISO to 6400 at the beginning of the render (found out the hard way it doesn't work to boost it when the render is half completed ) you can hit insane speeds and rendering glass and water will be a breeze

Hth

[fuzzy70]

5 Days

Honestly I would give up 3d for life if I had to wait that long for a render lol.

I have to be honest & say that I've not used Lux for a little while for my renders. Not because of Reality as that does pretty much everything I need but Lux, i'm still peeved that there is still so much of "Classic" not implemented into "Core" & instead get "Shadow Catchers" etc. I believe in the "get everything working before adding new features" train of thought.

I admit the power strategy did help on some of my renders but still there is some noise issues, quite often I got an image that was cleaning up nicely then around the 1 or 2 Ks/p stage noise & red/white dots started to appear. That totally baffled me as i'm of the mindset that noise disappears with time & not gets introduced with time. If I make a "Metal Heavy" scene or use a lot of LEM's they are the biggest offenders.

As a test one particular problematic scene with a lot of LEM's in it (well really it is a single LEM that's spread on numerous objects) which I asked advice on here I redone the mats but used Iray, also I exported the scene & loaded it into C4D & redone the mats for V-Ray. I actually got cleaner renders from both than I did with Lux in less time, & i'm including the time it took to re-assign the mats in BOTH not just each ones render. I've even had cleaner renders from Classic on occasion in less time than Core accelerated modes. BTW to keep the time comparisons fair I deliberately made sure only my CPU was being used for the renders.

Again this is not a dig at Reality at all as like I said it does it's job very well for me, even more so with the new material copy/paste feature.

I tried to register on the Lux forums but it wouldn't let me because of my IP or ISP was on the banned list?, never had that problem before & have had no problems joining new groups or forums since. That was pretty much my "Stuff you Lux" response, albeit not so family friendly as that.

The current types of scenes that have taken my interest of late are dark & pretty much Low Key style & have been unable to get the results I require with Lux using acceleration so I have to think hard about my workflow, such as do I waste time seeing if the scene will render fine with LuxCore, Just stick with LuxClassic or go the Iray/C4D V-Ray route.

Regards

Lee

[Marcel2586]

But it is worth the waiting, very nice Bandid!

[banditcameraman]

Hi Lee,

I'm sorry to hear about your problem with registering on Lux I have problems like that sometimes because of where I live

And the 5 days was to hit 105 K s/p; normally overnight will do and I usually let it run to 20-30 K s/p

And it's funny that you mention low key lighting, I did a render which is more than low key, it's tenebrism. It's also an instance in which the Power Light Strategy did not yield the desired results, probably because the lights in the scene needed to be sampled equally (I think )

Here it is





The light comes from the priestesses' hands, which is DzFire's torch construction kit. I turned the fires into emitters and scaled them down in Lux. It's rendered with CPU-Acceleration and ran overnight. Am pretty happy with it

[banditcameraman]

[owakulukem]