FurryBall 4 - new speed compare (unbiased rendering) V-RAY, Iray, Mental ray, Arnold


#61

This is what I call quality thread derailment :smiley:


#62

4k - 4min render (gtx460) sss shader (no displacement)

model downloaded from http://www.ten24.info/?p=1164

direct link
http://img571.imageshack.us/img571/6295/f49h.jpg
recommended to watch in 1:1 :slight_smile:

p.s. turn table 720p

//youtu.be/PYO5bNhY2d4

1frame ~ 13sec
200frames ~ 40min total

cheers


#63

Double post


#64

Looks like a very high poly head mesh being rendered with sss material and high res surface mimic textures and lit with an hdr map at 4K on an old gfx card to me.

Thx for the vid. Looks nice


#65

yes, it is a highpoly mesh with hires tex, there is nothing to hide here just a presentation of sss material…

real test is coming up, 99% in the morning, with moving cam, moving obj & motion blur…


#66

Can’t wait to see it. Please add also some deformation moblur into the mix if you can


#67

Actually 1 hour and 20 something minutes.
Sorry, rendered sequences with wrong version and now rendering again bazuka’s scene.
27 sec per frame, 200 frames, should be fast :wink:


#68

https://www.dropbox.com/s/km9ej4vwpkerkzz/RSrenderMBOff.mp4

MB off, had some issues with baked parts of blackboard, bazuka will fix them but in the mean time… avg render time 16sec per frame.
There is some flicker but would need someone that actually knows how to optimize render :slight_smile:


#69

mirko was faster, here is mine (should be the same :slight_smile: ) rendered with gtx460

//youtu.be/FqntjtMs-SU

Rendering frame time: ~74.572 s
Rendering done - total time for 200 frames: 20621.9 s

like mirko already said there are some probl with motionblur on shattered object and dynamics, also there is a small GI flicker as you may see, as i promise ill will try to clean up everything untill we get a clean no GI flicker render :slight_smile:

cheers


#70

but this is turning into Redshift forum so I suggest to stop here now.
I would love to see render times with Furryball and no hair as those would give closer look to comparison, at least till RS includes Hair rendering.


#71

Thx for the vids guys. The dynamic sim is too slow to judge about MB which maybe contributes to the scene looking too cg?.. But yep, a dedicated redshift thread would be more adequate to discuss. Cheers


#72

if you have time, im not an animator, i would like to render some scene if you have as a test

cheers


#73

While this thread has somewhat sidetracked its interesting none the less. However the biggest inherent problem with gpu rendering is memory. The Titan is limited by 6GB of ram, also what is the ability of the gpu to run custom shaders? I can’t see cpu’s being replaced by GPU’s in rendering anytime soon. Until GPU’s have the ability to use off card memory, or the cards themselves have lots of ram.

However i am biased myself, i’ve fallen in love with Arnold, and compared to all other renders blows them out the water.


#74

Btw this topic was at the beginning comparation FurryBall to other commercially released renders, we used for testing.

At the beginning there was many words about our ā€œridiculousā€ times, but I didn’t seen any your results from renders and versions we submitted in comparation. (Image and hardware configuration).
Thanks

This was our beginning:
http://aaa-studio.cz/FurryBall/FurryBall4compare.jpg


#75

Well one of the main reasons nobody bothered to run various tests was that you wouldn’t answer our questions. The second main reason would be that Redshift, while having a dubious name, is obviously a more potent contender for you to compare against.

The reason I asked you what a photon was is simply because mainstream physics cannot answer this question. I wasn’t trying to make you feel stupid - Maxwell and Rayleigh and Laplace couldn’t answer this question. Einstein couldn’t, nor Bohr, nor the rest. This is actually a huge factor in your rendertimes, even moreso in other renders though. There’s no such thing as an unbiased render - because they’re all using bad, squishy math at their cores.

Because you have to fudge the math to get results even approximating what our eyes see, the processor (GPU in your case, CPU in most others) has to do a ton of extra math to compensate. Rayleigh scattering itself is total hogwash. Almost all of the light-transfer equations have been rewritten to conform to Unified Physics, and yet you don’t see any of that converging on rendering technology yet.

Redshift itself has been falsified since the 1960s, and yet here we are still worshiping one of the greatest errors in physics. But when it comes to rendering tech, you’re going to have a serious contender with Redshift, my friend. You can go ahead and leave other renderers out of your comparisons at this point, and focus on catching up to Redshift. Granted, Furryball has great features and is obviously an excellent renderer!

So regardless of my naysaying, good work on Furryball!


#76

InfernalDarkness> What answers?
Just download our scene (In one pack there is Maya and Max) and try to make better time in your favourite listed renderer. If you want to see our setting for Mental Ray - Max scene is tuned for Mental Ray. We can’t compete with unreleased renderers, so please try this listed and already released renderers.

Yes, you are right we are not the best with other renderes, so we will be happy with your times.

http://furryball.aaa-studio.cz/aboutFurryBall/compare.html


#77

Don’t worry about my questions, they should be irrelevant to your work. The CG world is not yet ready for real chromatics, photometrics, and physics, including photons - and since you ignored my question so blatantly, neither are you.


#78

nice tests… but what was CPU configuration for redshift and vray renders? I cannot find it…


#79

I tried out the scene in Maya with Redshifts latets beta build. The scene only had furryball materials so I had to redo all of them the best I could. I used the eariler renders as reference and tried to guess the settings the best I could. I used Redshifts own materials, sky simulation and bokeh lens shader. Redshift doesn’t support fur yet so instead I used a displacement material with a lot of pointy ends to compensate. I used an overclocked GTX 780 for the renders. I did 3 versions with different settings.

Here are the results:

Irradiance Cache + Irradiance Pointcloud (low settings)
http://www.youtube.com/watch?v=tVJxJZUxiTg

32 seconds

Brute Force + Irradiance Pointcloud (low settings)
http://www.youtube.com/watch?v=IsLhk5Zvz_4

1 minute 5 seconds

Brute Force + Irradiance Pointcloud (high settings)
http://www.youtube.com/watch?v=eSkJw26rNww

2 minutes 20 seconds


#80

And that same scene and image, highest settings rendered with 4 titans

INFO : [Redshift] Redshift for Softimage 2014 SP2

INFO : [Redshift] Version 0.3.39 Beta, Feb 7 2014

INFO : [Redshift] Rendering frame 1

INFO : [Redshift] Scene extraction time: 0.028 s

INFO : [Redshift] Rendering time: 72.843 s (4 GPU(s) used)

INFO : [Redshift] Rendering done - total time for frame 1: 72.879 s