V-Ray Scene Optimisation Script

[Morne]

Well, I cant say I understand your script, at least from the math point of view, but having now played a lot with it, and experiment myself without the scripts help, I have got my BF/LC 6k still renders out in just under half the time, 2 hours down to 1:05. Very pleased! Fair bit of planting and large building, CG in about 60% of images, rest BG photo.

I'm not using all your suggestions such as untick the magic divide shading subdivs, but I have a workflow where the 1h (cleaner then before as well!) render can be a 6 min full size draft with just two number changes.

Time will tell if my "standardized" settings will work out, but they are better then before and I have a good understanding of the reasons behind the speed up, so many thanks. Now holiday time for 11 days as been solid for the last 4 weeks!

M

You cant say things like that and not share what you did to get to that
Enjoy your holiday

[m_hinks]

Enjoy I did! Now to deal with the 600+ emails!!

I'm still testing but have now also got an 18k x 4k pano render down from 10 hours to 5 and cleaner! This is fun, and I got a couple new blades coming tomorrow!!

[rajdarge]

I'm not sure what happened but a render at 22Mpx with 15M particles just wouldn't render. I even tried switching from BF to IRmap. The BF times were 360h to render, and 133h with IRmap. I noticed also where there was Transmap hair, the render buckets would blow out where it was present in the scene. Is there some reason that semitransparent objects stall the renderer after the script is applied?
Apart from that its fantastic to get lovely clean shadows with a minimum of guesswork. Thanks so much for this.

[^Lele^]

embree is turned on.
Try turning it off, i have no idea if it's meant to work with hair in the version you're using.
Other than that, i can't think of anything else off the top of my head.
Send me a stripped version before and after, and i may try and figure it out.

[rajdarge]

after a lot of fiddling, it was the embree that was killing the render time. turning it off helped a lot.
Why does checking "divide shading subdivs" render faster, but with a huge increment in noise?

[^Lele^]

read the thread again from the start

[LudvikKoutny]

Since i found those ridiculous rendertimes in this thread quite funny, i decided to do some testing myself.

I took my original Corona scene, and converted it to Vray as best as i could. Only real difference is i did not convert book cover materials because there was lots of them, so i replaced it by brown slightly reflective material.

So, here's original Corona render that took 7 minutes and 12 seconds on my PC:


Then i have set up Vray scene to match rendering techniques used by Corona, so i have used Brute Force as primary and Light Cache as secondary GI. I have set up sampling to be as optimal as i could. Rendertime of Vray scene was very similar, clocking at 7 minutes and 45 seconds. Some areas are slightly more noisy, and some slightly less:


Then i have resaved the scene, and tried to use all techniques Vray has at it's disposal, to get best quality/speed ratio. I used irradiance cache with detail enhancement, light cache for glossy reflections, stored direct lighting within irradinace cache and so on. Resulting quality was slightly worse, but rendertime was 2 minutes and 3 seconds on my PC. Here's the result:


Please keep in mind i have very old PC (Core i7 870 - one of the very first Core i7's ever) so on today's mainstream haswell processor, rendertimes should be a lot shorter. I would guess around 50 seconds for Irradiance Cache version of the scene, and about 3 to 3.5 minutes for Brute Force + LC version.

I am attaching both Vray versions of the scene, so you all can do your own tests:
https://dl.dropboxusercontent.com/u/...ampleScene.zip

[^Lele^]

I'm glad you're amused.
Yeah, I'd use IRMap.

[eyepiz]

Script updated to 1.50 (in the first post).
Let me know if it sorts out the oddity, Alex!

Ah, Eyepiz, i meant for you ro tender it with the offline (ie, standard, CPU) renderer...

Hi Lele!

I know it's been a while but I ran the vr_scene06.max on a 28 Core machine I'm working. I tried to test the new RT GPU LC but it crashes.

[3LP]

Hey guys,
I thought I would give this one a go and found why RT GPU was crashing.
This scene has been converted with a script from corona to Vray I'm guessing so it messed up with few things and the one that made it crash was the second domelight called Dome_under.

After doing a bit of cleaning, here are the results :

Linear :


With Reinhard .5 applied :


It's not exactly the same look as your previous renders but with reinhard applied at .5 it seems to be closer so I saved a version.


Here is the scene cleaned up id anybody would like to give it a go:
https://dl.dropboxusercontent.com/u/...6_RT_Fixed.max

Rendered on 3x GTX 970 with max ppp at 8192.

Cheers

[3LP]

I thought I would give it a go with the adaptive settings as this might make more sense then just a max ppp.

1min :


2min :


3min :


4min :


6min (Noise .002) :


scene :
https://dl.dropboxusercontent.com/u/..._Fixed_V02.max

3x GTX970

Stan

[savage309]

I had some time so I decided to run the test with a nightly build and one Titan X.
1m.

2m.

3.5m.

4m.

6m.

Btw, the build these are made with contains some optimizations + QMC. With coherent tracing on it is 10-20% faster.

[^Lele^]

Nice Tests!
Mind sending me (here, privately, as you please.) the EXRs for the images, so do do a noise level check?
That, or ask vlado for the nuke gizmo i gave him.
I'm asking so i can set up a CPU renderer target which matches the final amount of sampling, for direct comparison.
EDIT: final amount of sampling with lower camera rays shot, that is. i can read the KPaths from the images

[3LP]

We are building a 7 Titan X render-node for the moment.
Can't wait to play with it, it might be able to knock out 1080p frames under 5 min and hopefully with SP2 it will shine.

I'll give this scene a second go once it's up and running.

Stan