Benchmarking comparison of Iray, Lux and keyshot

DrunkMonkeyProductions

Since I finally got some hardware i consider up to the task of using iray i decided to do a bit of benchmarking to see how it measures up to lux and keyshot.

How the tests were done
SOFTWARE
Iray(daz studio 4.8.059)
Render settings:default
Time was adjusted to achieve 95% convergence.

Lux(1.5.1 build 15282)
Reality plugin(4.2.1.16017)
only the skin shader of g3f was changed to the Reality preset.
The HDRI from iray was applied as an IBL light.

Keyshot(5.3.6)
G3f exported as OBJ.
Default human skin 2 used for skin. Missing textures reapplied, and HDRI loaded into environment lighting.

Each render engine was left at the default load render settings, and no adjustments to anything were made in the scene.

Due to the default time settings of iray being fairly low, the time had to be adjusted to achieve 95% convergence on the workstation as it would exceed the limit of 2 hours.
This adjustment was done due to the constant focus of users on the convergence percentage in relation to renders.

HARDWARE
Workstation
Dell precision 690
Xeon 5130, 2.0GHZ. dual processor
16GB ram
Nvidia fx quadro 1800, 786MB(pcie x16)

Server
HP proliant DL 385 Generation 7
Opteron 6140, 2.6ghz, dual processor
12GB ram
ATI FireMV 2250, 256mb(pcie x1)

ASSETS
SINgle character scene
G3f default
IG IRay preset used for iray renders
Reality preset used for Lux renders
Keyshot Translucent skin with G3f textures applied.

Swimsuit for G3f(22886)
Preset 15

Hannah hair
Default for all three render engines

CHARACTER AND SET
G3f default
IG IRay preset used for iray renders
Reality preset used for Lux renders
keyshot Translucent skin with G3f textures applied.

Swimsuit for G3f(22886)
Preset 15

Hannah hair
Default for all three render engines

Classic Deco( )

TESTING
Each test was run on both the workstation as well as the single server.

IRAY Notes
For iray the tests were adjusted in time only to achieve 95% convergence. This was done due to the focus of users on this number.
Classic deco shaders were converted to IRay default, except window which was converted to iray glass, solid clear.

Keyshot Notes
Due to a difference in orientation in keyshot the HDRI loaded at about 90 degrees off from the position in iray and lux. This was not adjusted to maintain the "default" settings.
Keyshot, while a PBR, is a biased render, as such it simply completes as opposed to using samples or Iterations as a measure of completion.

Results

Single character scene
Workstation
Iray

Render time:28 minutes 4.44 seconds
Convergence:95%(default)
Samples/iterations:432

LUX
Render time:18 minutes 47 seconds
Samples/pixel:445

Keyshot
Render time:29 minutes 52 seconds

Ranking(highest time to lowest)
Workstation:
KS    29 minutes 52 seconds
Iray:    28 minutes 4 seconds
Lux:     18 minutes 47 seconds

Server
Iray

Render time: 5 min 22.64 seconds
Convergence 95%
Samples/iterations:431

Keyshot

Render time: 6 min 25 seconds

LUx render

Render time:3 minutes 49 seconds
Samples/pixel:434

server
KS:    6 minutes 25 seconds
IRAY:    5 minutes 23 seconds
LUX:    3 minutes 49 seconds

Character and room

Iray render

worstation
Render time: 6hrs 34 minutes 50 seconds
COnvergence:95%
Samples/iterations:2057

LUX
Render time:6 hrs 40minutes
Samples/pixel:2060

keyshot
Lux:    400 minutes 22 seconds
Iray:     394 minutes 50 seconds
KS:    19 minutes 59 seconds

 

Server
IRAY
Render time:1 hr 33.39 seconds
Convergence:95%
Samples/iterations:2069

LUX
Render time:1 hr 5 minutes 13 seconds
Samples/pixel:2060

Keyshot
Render time:4 minutes 17 seconds

Lux:    65 minutes 13 seconds
Iray:    60 minutes 33 seconds
KS:    4 minutes  17 seconds

 

Conclusion.
In one situation Lux performed better than iray, in another it performed worse, Keyshot was the same.

 

Personal thoughts

People are hung up on the render times associated with iray and blaming the software, while ignoring the two biggest factors.
1. the user
2. the hardware.

The user, in many cases, hasn't been working with iray long enough to understand it's intricacies. I include myself in this list.
Just like 3delight and lux, we weren't getting good results and fast renders right at the start.
When i started with 3delight, my renders averaged between 8 and 10 hours. Those same renders today will take about 1-2 hours to render.
The difference?
I've studied the various settings of 3delight, shader settings, and lighting.
This took a lot of time, in both work and renders.
I've also upgraded my hardware.

The Hardware.
IN 3d graphics the more processing power you have, the faster the renders get done.
Most hobbiests don't think they have the budget, or the space, to setup a proper workstation and/or render farm.
They also don't know enough about hardware to make informed decisions.

so a quick run down on hardware considerations.

FIrst thing: BUy used, you'll get a lot more for the same amount as new.

1. MOtherboard: Get a system with at least two cpus, and preferably 4.

If you plan on GPU rendering you'll need to consider a board with at least two, and preferably more pci-e x16 slots.
Be sure to check how they are wired(x16, x8, x4) and try to get boards where all the slots are x16 if possible. Many will drop performance, down to x4, if all the slots are used. this can result in a performance drop of upto 60% or more effectively negating the additional cards.

2. Memory. The more the better. The type and speed are going to be fairly irrelevant. The minimum i recommend is 16GB and preferably 32 or more.

3. Graphics. For non-gpu rendering, it's fairly irrelevant. The problem with low-end graphics cards is that it'll be hard to work as the GPU won't be able to update the screen fast enough.
SO at least 1Gb ram and preferable 4 or more.
For GPU rendering, one for the monitors, and at least one for rendering, don't try to use one for both. 6Gb minimum and as many cuda cores as you can afford.
Preferably in the Tesla type, or at least workstation class, as these are geared for number crunching and extended usage.

4. CPUs. Really it's irrelevant whether you use Intel or AMD.

The only considerations are number of cores and speed.

One last thing to consider if you plan on GPU rendering are adapters for the smaller pci-e slots that allow connection to an x16 card. While you'll have a performance drop, they may be a viable option depending on your situation.

GPU render boxes. There are external boxes available that allow the connection, generally by usb or adapter card, to connect additional gpus to a computer.
The biggest i've seen handles 16 tesla cards.

The draw back is the cost. Even on the low end these boxes cost more than buying another used computer.

 

A few thoughts about the confusion in rendering with IRAY(or any unbiased render engine).

1. They never complete they are only stopped. Whether this is done manually or through settings.

2. The Convergence render setting, what the progress bar shows, is not a measure of completion in and of itself, it is a measure in relation to other render settings. Stop thinking it's a value of completion it is not. It is not the same thing as the progress bar percentage in 3delight, that one is a measure of completion as 3delight is a biased render engine, just like keyshot.

3. Higher iterations/samples does not always mean a better render. While the renders in this test could easily be run longer, there is no guarantee that the increased samples/iterations will result in any improvement of the render quality. Lighting, shaders and other render settings may be the cause of the grain as opposed to just the iterations/samples.

4. There's no hard and fast equation to a good render. As demonstrated in these tests there are too many variables even in a simple scene to say do XYZ to get a good render.

 

jeffglobalmedic

I used Keyshot for a little while until I held my nose and learned Marmoset Teabag2.  Either of them meh, but Keyshot is kinda a toy.

"There's no hard and fast equation to a good render. As demonstrated in these tests there are too many variables even in a simple scene to say do XYZ to get a good render."

That's actually interesting.  I've been torturing myself with V-Ray tutorials and it may have served a useful purpose.  Idk y it took me until yesterday to realize it's refraction that makes water and glass, water and glass.  Hmmm.  Spikes rendertimes though.  Does iRay have a rendersample pass?  Well, do they have any multipass (man that Ukrainian actress really messed me up, Fifth Element)?  Cause if the end user doesn't get ANY feedback on what's going on in the engine itself, we r screwed and are more alchemists than practicioners...

Anyone can tell, well most anyone, if they see red pixels in their rendersample passes.  There are primary rays for lighting and "secondary" rays for shadows, for example and the way the ray tracing engine uses primary and secondary multiples to determine number of light bounces was weird to learn for me.   With the rendersample pass, if the scene is too red, increase samples (simple explanation).     If it's almost all blue, you just spent the summer rendering 2 frames if you're not rendering with a render farm.  I think my render farm really sucks, it's one workstation and the old hampster is almost in mutany mode...so I think I'll start a render and then check it in a day or two if it didn't make an escape. On the upside, my bro will give me his old graphics card which is MY graphics card I use now, so I can SLI at least and heat my room to tropical temps this spring and summer.

And I don't think you mentioned, nVidia GPUs or you're screwed either.  So maybe the CPU pedigree doesn't matter, but the GPU's pedigree is all important.  I wonder what percentage of the population even makes a distinction of GPU and CPU?  I saw a video with Korean girls shown most of the current female western pop stars and they all said, "They all look alike!" 

I'm here for the iRay renderer from iClone.  So far in the 5 or so days I've been playing with iClone and the day and a half I've been playing with Daz Studio, I can export and import between the two, but I can't find a model that has her face boned yet.  Very weird.  I just started the 3Dxchange with a Genesis 2 model, so I assume, duh.  I was on the forums looking to find out which models can morph their facial features and lipsync given facial bones for the purpose when I saw this article. 

 

I've seen some of your youtubes and do appreciate them.

 

 

prixat

Hi jeffglobalmedic

Un-biased renderers don't need a rendersample pass, since they continue rendering to an infinite number of samples. 

Kevin Sanderson

How to render faster? Watch this:

 

And jeffgloablmedic... G3 has face bones now. G2 didn't.

jeffglobalmedic

@Kevin Well now, that makes sense.  Maybe my 2 day self amassed primer missed something and my 20/80 idea, that if you know 20% of something you can do 80% of what you need has failed.  Today, I'm just checking out a FRAME rendering in iRay now 15 mins in...uh oh.  Well I'm downloading Daz on my workstation now, cause obviously my laptop is just getting murdered.  Took 20mins for a frame, yay, will definately look at that tut, ty.   [Edit]  I did see that tut before, I'm rewatching it to see if it just went in and out without a trace...  I will also go down the list of stated tuts in the forums as well.

...and, is the Genesis 3 ppl free, but with no clothes or anything, cause it looks like I have it, but I don't have anything really for it...

I don't even know if I can import the audio for the song or I have to use Premiere to sync up the movements to the sound like they do with foley tracks...

I apologize but I just got the program like 4 days ago because of the iRay renderer after iClone gave me such a poor render, I just said WTF!?  To be fair, I didn't invest in the Indigo renderer because so far ppl say it's buggy as hell.

So, looks like structured learning for a few days to not miss stuff and more quickly get a trick set to save time, and produce good results.

@prixat But that's the point.  You can at least in VRay in Maya and Max limit the potentially infinite time to convergence, but how do you know how much to give the renderer?  I was shown samplerender passes to gauge a frame and then let 'er rip.  Doing 10 sample renders increasing your sample size after turning off quality until you get the right amount of samples would suck if I had to do it that way just with my eye.  So I put it at .25 quality and it took 20mins on my laptop.  She's looking at me meanly right now, saying no way is she gonna do that again!  I told her she is just lazy, and not to worry, I'll be cooking my room with the workstation from now on.  I rerendered the scene with 250 samples too and it too, oh no, the window auto closes.  It was less than 10 mins, and the quality was hit hard.  Wish we had a way to use a render buffer like in Maya, so I could just move back and forth, but I used paint.net instead of firing up PS and it taking over my screen.  A comparison is attached.

I got lucky and also saw this video, which I remember the author is ridiculously good (and he has 2 (two) TitanX's) mofo!  If I understood him correctly, using a small iRay render window, eliminated the gpu having to have each frame be loaded into it's memory saving almost a minute a frame!  Well that adds up!

 

Ok, got the Daz Studio software on my workstation now and he's saying he's ready to comparison render...whoa, I have it set to the 0.25 quality default samples and the workstation is making my laptop with it's silly mobile gtx610m or something look like counting on fingers, but that was only after I manually turned on the GPU, the software is "dumb" not turning on the gpu by default. Wtf.

Total time from 20mins on laptop to 5:38 over 580 samples.  Idt the having iRay in a tiny render window speeded up anything but maybe 50 secs, cause of preload.  Gonna render the mute puppet sequence now just to show the iClone forum ppl that it doesn't import back a jumbled mess.

 

enjoy