V-RAY FOR SOFTIMAGE BETA - finaly here

[Zafar Iqbal]

Good idea. I must do that. I've set up a depth of field test so it will have to be after that.

I agree with the funny business of Unified Sampling - my experiences with it are very close to how you describe yours.

[Zafar Iqbal]

I don't remember the settings too well but the scene/settings is only slightly different from the previous one. mental ray max samples were upped from 6 to 8 and cuttoff was lowered from 0,005 to 0,002. Thats the only thing I remember for sure.

mental ray, 7min 16sec


VRay, 7 min 46 secs


VRay is slower and looks "much" worse - especially if you plan on doing some post work.

It's interesting that the visual/shading difference is much larger in this test.

[Memag]

Great tests,thanks. Can you throw in 3delight as well?

[Zafar Iqbal]

Can you throw in 3delight as well?

I thought about it but dropped the idea since the freebee only uses 2 cores, while mr and vray are using all 6 that I have. I guess I could get theoretical times by dividing by 3.

[Zafar Iqbal]

This is... odd

VRay, 3 hours 10 mins
Adaptice DMC
- min 1
- max 4
- threshold 0,005
Area filter @ 1,5
Brute force GI on prim and secondary with 44 subdivisions. 3 bounces
DoF @ 32 subdivisions
All walker models are VRMeshes

Noise is very acceptable and holds up for post work.



mental ray, 9 hours 5 mins to render 40%. Cant be buggered to let it continue rendering.
Unified Sampling (better sampling than before)
- min 1
- max 8
- quality 0,9
- cutoff 0,002
Triangle filter @ 2
Brute force final gather @ 384 rays. 3 bounces.
Bookeh shader with 24 samples
1 real model and 9 instances.

Quality is worse than that of the VRay render.



!!!

Someone tell me I did something bad. This is just horrible. I must try again using classic AA.

[Memag]

I really can't see substantial quality loss in MR images. But I am a bit blind these days.
Time it took to "render" the image , however, is big factor is it not?

I wonder how long it would take for 3delight to render it with 2 cores, and how would final rendering look like?

[Zafar Iqbal]

I eyeballed the quality when I was setting the scene up and you are correct in that it's difficult to tell which is better, but it's important to pixel peep in order to get proper comparisons.

I will do a 3delight test shortly (hopefully) - I'm just in the middle of some other things right now. I'm not too used to 3delight, but it wouldn't surprise me at all that 3delight would blow both mr and vray way off

[Zafar Iqbal]

I'll have to revisit 3delight some other time. It's not entirely behaving as expected. I get very long "pauses" before rendering starts and even then it halts shortly after starting.. and then resumes rendering. When I finally thought I got my head wrapped around things, I went for a render of 10xAT-AT walkers and ended up using up all of my 20GB mem. The original scene is an old one from when I previously had a go with 3delight and I had 8GB back then - so I'm not sure why. The PC is swapping like hell but I managed to abort and enter render settings. No clue whats wrong. I don't see anything that should be affecting memory like this.

[McNistor]

I'm not sure as I didn't got the chance to use unified sampling in a serious render and neither have I studied the docu thoroughly, but I believe that cutoff is a threshold value and setting it to .002 you give it very little to none room to be adaptive. I would increase it along with max samples and let MR decide where it needs to add more samples. You could try .1 for cutoff and 32 or 64 for max.
Just an idea.

[Zafar Iqbal]

I will try that. The cutoff parameter makes sense and works as expected but I have no idea of what Quality does... controls quality.. yah, gaaaah Would make more sense if it had a fixed range, but it doesn't.

[ActionArt]

Ya, typical mental ray! Take a whole year to introduce a pathetic new feature that is slower, more confusing, worse quality and even more stupidly named than the previous one.

"Quality" from 0 to 1 but hey, you can set it higher than 1 ??? So what does that do?

Cutoff - cutoff what exactly?

I've tried a lot of settings and the results are dismal. I never use it anymore. It seems to get into feedback loops or something where the render will just drag in certain areas, especially with lots of reflection. The results are crap too, even with motion blur which it was supposed to help.

It seems we're not alone either, lots of others complaining as well.

[Zafar Iqbal]

I got the mr render down to 2h 40 mins however I have reflections on the floor. They must have slipped in while trying to set up for 3delight.

I landed on 16/64 and cutoff at 0,1. DoF sampling on 6

I will try again and this time do similar optimizations to VRay as well.

[origin]

You can speed up vray's brute force gi ALOT by setting secondary gi bounces to lightcache. With new retrace option there should not be any flickering as long samples are set high. I've renderer some commrecials with this technique and it works very nice.

http://www.spot3d.com/vray/help/200R1/render_params_lightmap.htm
http://www.spot3d.com/vray/help/200R1/examples_lightcache.htm

[Zafar Iqbal]

origin: thanks, I might try that at some point, however for now it would beat the purpose of a direct comparison.

A followup to my recent test.

mental ray, 2h 10mins
- min 16
- max 64
- quality 0,9
- cutoff 0,1
Triangle filter @ 2
Brute force final gather @ 384 rays. 3 bounces.
Bookeh shader with 8 samples
1 real model and 9 instances.



Seems like I should have gone with the 32/64 sampling as I was suggested This render is has more contrast than before. I must have broken something while I was trying to do a setup for 3delight.

VRay, 3h 10 mins (re-post of the previous render for convenience)



I'm open to suggestions on how to improve VRay sampling settings. I tried but didn't really succeed.

[Memag]

I really like the higher contrast, and deeper gray of the MR renderings.

[Zafar Iqbal]

I really like the higher contrast, and deeper gray of the MR renderings.

I solved the puzzle about why the render had more contrast. I used 8 FG rays and not 384 as stated (I've corrected the post now). I guess this also mean that, even though I like the contrast myself, the render is less accurate. I'll have to redo the test with enough FG rays.