PDA

View Full Version : rendertime ok?



madno
03-29-2016, 09:30 AM
Hi,
found a thread from another user who asked about surfacing his oxygen mask. I played around with it and now am wondering if around 14 minutes render time is ok for the attached render?

133166

BeeVee
03-29-2016, 09:49 AM
We need more info. Was that rendertime for that exact render (1280x1000) or was it bigger? What spec machine do you have? What version of LightWave are you using?

B

Danner
03-29-2016, 11:39 AM
Blurry refractions are a huge render time killer, so.. no, not surprised, actually it's not bad at all. If it's a still you could use the interpolated feature in dielectric, it flickers in animation but looks ok for stills.

madno
03-29-2016, 11:31 PM
Machine is a dual xeon (a few years old).
Size was 1280x1000 - no resize.
LW 2015.3 with HDR as lighting (put into textured environment and rendered with importance sampling). No LW lights are used.
Materials are delta and sigma2 for the mask and procedural (IFW2) for the table.
The "noise" in the clear part of the mask is due to bump by a turbulance node (I liked the bumpy look more than the pure noise free reflection blur).
The mask had reflection and refraction blur active.
Ray passes set to 12
AS set to 60
GI set to primary 128, secondary 32

Tried interpolated blur now, but it does not like the scene. Setting sample and blend to 1 gave me this after 6 min 39 sec.:

133172

spherical
03-29-2016, 11:58 PM
AS set to 60

Really? Why?

kolby
03-30-2016, 03:50 AM
Hi,
found a thread from another user who asked about surfacing his oxygen mask. I played around with it and now am wondering if around 14 minutes render time is ok for the attached render?

133166

Hi, that thread was started by me.
I used Delta material and classic surfacing with different IOR and blurry reflections and refractions.

Render settings:
1920x1080, non interpolated MC, 1 bounce, RPE 16, Shading samples 8, Light samples 1, Ray recursion limit 8, Ray precision 5, Ray Cutoff 0.03, HDRI lighting, ISBG samples 256, no lights

AA:
Min samples 3, Max samples 32, AS Threshold 0.02

No ground object, just the mask.

My render times on i7 920 (2.8GHz) was 20 to 45 min./frame, depending how close the mask was to the camera and there was noticeable level of noise which I removed in Fusion. (Neat Video denoiser doing great job)

Now you can compare :)

JonW
03-30-2016, 06:33 AM
Ray recursion limit, you only need 1 higher than the number of transparent surfaces aligning in the render. 3 should be plenty for most things.

Upping Ray cutoff a bit to say 0.05 should also help a bit also without noticeable affect.

madno
03-31-2016, 12:16 AM
Hi Kolby,
thanks for chiming in. I also tried with non interpolated MC. But like you I got a lot of noise (did not try to remove it in post), so I gave up on that.
Knowing your render time now, I think mine is ok. But Sphericals hint just reduced it by half !

To Spherical
"Why max samples set to 60" - good question. I lowered it to 30 now and still get a good result (6 min 57 sec. :)

133184

To JonW,
yep normally I reduce Ray recursion but in this case I liked the subtlety of more subsequent reflections.

133185

And here is another one with changed surface settings (7 min this time)

133186

Thanks for your remarks and
Kolby, thanks for the model to play with.

spherical
03-31-2016, 12:52 AM
Try lowering Max Samples to 12 or 9, Threshold to 0.07 and compare render quality.

madno
03-31-2016, 01:37 AM
Another good hint. Now I am at 3 min 53 sec. That includes 2 min 23 sec. for the sigma 2 preprocess.

Somehow I thought, I had the experiance that LW needs a lot of samples in cases like that (bump and blurred reflection). But that does not seem to be the case.

Below is a combination of two images. One was rendered with 30 samples and 0.01 Threshold, the other with 12 / 0.07.

133188

Zoom:

133189

madno
03-31-2016, 02:06 AM
And here is another comparison:

Left is an 8 min something render (samples 40 / threshold 0.06) right is the 3 min one (samples 12 / threshold 0.07).
The 8 min one looks better for sure, but the other one is not to much worse, I think.

133190

spherical
03-31-2016, 03:13 AM
Surely not enough difference that, without a direct same time comparison like this, one would ever determine that the 12/0.07 render was lacking.

JonW
03-31-2016, 04:12 AM
I usually have about 12 to 16 for minimum samples & maximum samples about 2 to 3 x the minimum.

Depending on the surface texture or what is being reflected in it, having too low Minimum & too high Maximum quite often takes longer to render than having a reasonably high Minimum, which in turn speeds up the Maximum.

Reconstruction filter Gaussian
Sampling Pattern Classic

kolby
03-31-2016, 04:57 AM
... Kolby, thanks for the model to play with.

You're welcome :)

vonpietro
03-31-2016, 07:55 PM
BY THE WAY what did you set your lights and shading samples to... default is 1, however when you increase them, render times do go up, sometimes quite a bit depending on how many area lights you have.

jasonwestmas
03-31-2016, 08:10 PM
Somehow I thought, I had the experiance that LW needs a lot of samples in cases like that (bump and blurred reflection).

133188

Zoom:

133189

We used to need a lot of AA passes with the classic camera, but not necessarily with today's AS samples in LW. It got overhauled quite a bit.

madno
03-31-2016, 11:17 PM
BY THE WAY what did you set your lights and shading samples to... default is 1, however when you increase them, render times do go up, sometimes quite a bit depending on how many area lights you have.

Both were at default 1

spherical
03-31-2016, 11:51 PM
I usually have about 12 to 16 for minimum samples & maximum samples about 2 to 3 x the minimum.

Depending on the surface texture or what is being reflected in it, having too low Minimum & too high Maximum quite often takes longer to render than having a reasonably high Minimum, which in turn speeds up the Maximum.

Well, that's interesting. I'm always ready to learn new techniques, but I considered the high Minimum method to be a false economy. To make sure I was not thinking incorrectly, I performed a test. Obviously, things like this are somewhat scene-dependent, but the overall principle is what is being tested. First, I never find the need to have Minimum Samples that high and certainly not Maximum Samples at 2x - 3x that level. Pretty much has been demonstrated as a diminishing returns overkill waste of time.

The scene has many reflectives, refractives, and many, many complex and also simple textures, blurred reflections, material nodes, Surface Editor gradients & bumps, you name it. 158 lights (not all in the one frame, obviously). 7.8M polygons. 55+ detailed models in the frame; 9 of them instanced. Ray Limit 16 to get all of the overlapped refractives; where 8 is normally sufficient to handle high poly count trees, etc. 1280 x 720. GI = Final Gather Interpolated, Directional Rays: on, Bounces: 2, RPE: 1000, Secondary Rays: 30

Minimum/Maximum Samples = 4/9:

Render Time Before Adaptive AA starts, including Pre-process & GI: 2h32m
Total Render Time: 3h55m58s

Minimum/Maximum Samples = 1/9:

Render Time Before Adaptive AA starts, including Pre-process & GI: 1h15m
Total Render Time: 3h37m0s

Final images are identical; even when flipping back and forth at 500% in Image Viewer or Photoshop.

Preloading AA into the render by upping the Minimum Samples does make the Adaptive AA portion shorter, but not only is it shifting a portion of the AA from one operation into another operation, it adds to the overall time; for no benefit. At least that's what works here. Standard disclaimers apply, YMMV, Professional Driver on Closed Course and all that. :D