Octane quick test-drive

Post your tutorials and help questions in this forum
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Octane quick test-drive

Post by thomas »

Hi there!

Since it's so quiet here I thought I would just post a very rough test of an old scene with Octane. No surface set.

For those that are curious: it's brute force rendering, so even though it's rendered on the GPU there are no miraculous speed gains. This took about 40 minutes on a GTX 760 (with 1152 CUDA cores), and it's still pretty noisy. More contrast in the materials will make it even noisier.

The material shading and surfacing though is amazing - some neat tricks in the workflow that make setting up physically accurate stuff supereasy.
Octane-Workshop.jpg
User avatar
Janusz Biela
Posts: 3265
Joined: Mon Mar 13, 2006 10:39 am
Location: Finland
Contact:

Re: Octane quick test-drive

Post by Janusz Biela »

To clarify: there is no difference in global illumination between Octane and Kray (or other software - all are based on the same GI mathematics solution). If in Kray you put Path Tracing render system you will get exactly same effect. Render must be done in pure Linear tone mapping without any processing. Perhaps you will get almost same effect also in QMC mode - not recognizable by amateurs .
If you can post this scene would be nice to test it.

Of course in Kray we do not have any shaders and this is big problem because final effect also depends from surface simulation.

The noise level in Path Tracing NEVER goes to be clean. Is just level which user accept.
Noise level in Path Tracing is exponential: next level of noise is longer then previous. Why renders in Octane, Arnold, Maxwell appear very fast in the beginning (big noise of course in first minutes) but later is getting slow with each step and to achieve quite clean render you have to wait many hours.

Probably in your scene you had only direct light. Adding another lights will increase noise and time calculations. Adding reflection/blur will make nightmare for waiting...

One Titanium Z GFX card allow render fast but to render comfortable you need 2xTitaniumZ + Powerful Power Supply. Also during rendering perhaps you have lags in operation system so you can do nothing during rendering :D
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Re: Octane quick test-drive

Post by thomas »

Hi Janusz,

Interesting to hear that all brute force GI uses the same algorithms - although it's to be expected: one year it's presented at Siggraph, and the next year all programmers use it!

I'll share the scene so we can see!

Also, good point on the exponential nature of path tracing noise. Marginal gains for longer render times... My scene only has the windows for lights in Octane - in Kray there are 20 lights in there, but Kray is a lot faster :D

By the way, in Octane for LW 2 there is a lag in the operating system, but in Octane for Lightwave 3 (open beta) they solved it, and the OS interface is really responsive, and I can do heavy work in Photoshop etc... while rendering with the GPU.

But in the end my big conclusion is: my dream setup is Kray, with the addition of some good new shaders. It's the shaders that make Octane great: they work well, are very predictable, and they are VERY easy to work with. There is only ONE way to make glass, but it's perfect. There's only ONE way to do SSS, but it's also close to perfect. The shaders just do what you want, out of the box, and tweaking them is predictable. If Gregor can crack this in Kray 3 (glass, velvet, anisotropic, fast and easy albedo setup) then we'll have a killer render engine.

Oh yeah, and subpixel displacement! Once you start using that there's no way back!
mdharrington
Posts: 33
Joined: Sun Sep 20, 2009 8:21 am

Re: Octane quick test-drive

Post by mdharrington »

The Octane PMC path tracing algorithm is slightly different from all others. The difference being that the algorithm fires samples towards problem areas and searches better paths to sample. Sort of like adaptive sampling or importance sampling. It takes longer but produces more accurate caustics then standard path tracing.

Other then that...a path tracer is a path tracer.

I would wager PT vs PT octane will be faster due to the GPU speed...provided you have a fast enough GPU.
Comparing PT to photon mapping is apples to oranges...and Kray should be way faster

Also remember Octane does have it's biased kernels....which would be similar to a final gather solution in a regular render, and they are much faster.

Knowing how to tune an Octane render makes a huge difference in time savings. Most people I see just use the defaults.
Coherent ratio for example can speed up the render by 40%....but flickers in animations. Also setting spec gloss and diffuse bounces can drastically speed things up. By default most are set to 24 bounces which can be overkill.
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Re: Octane quick test-drive

Post by thomas »

Alright!

It's good to see a bit of life in the forum again! It's a bit too quiet for my liking at the moment!!!

For those that want to play around with my scene - and get plenty of ready-to-use Dimensiva models for LW + Kray, here's the download:

http://we.tl/J70mMqaumu

Be quick, it's only live till january 28th! And let me know what you do with it!
ideart
Posts: 144
Joined: Mon Nov 16, 2009 4:34 pm

Re: Octane quick test-drive

Post by ideart »

Thomas thank you for the scene.
GTX 760 kills all the joy working with Octane.
I had to jump recently to Octane because I find Kray not reliable enough for animation and I wanted predictable results.
Sure render times are a bit higher and some serious investment in hardware is required but at the end of the day you hit render and you go to sleep with no worries.
Kray is great for quick stills and I am still going to use it for that but the BakeGI looping bug and the random freezes in time interpolation makes it very frustrating when you have to render 3000 frames.
sotto3d.com
User avatar
Janusz Biela
Posts: 3265
Joined: Mon Mar 13, 2006 10:39 am
Location: Finland
Contact:

Re: Octane quick test-drive

Post by Janusz Biela »

mdharrington wrote:The Octane PMC path tracing algorithm is slightly different from all others. The difference being that the algorithm fires samples towards problem areas and searches better paths to sample. Sort of like adaptive sampling or importance sampling. It takes longer but produces more accurate caustics then standard path tracing.

Other then that...a path tracer is a path tracer.

I would wager PT vs PT octane will be faster due to the GPU speed...provided you have a fast enough GPU.
Comparing PT to photon mapping is apples to oranges...and Kray should be way faster

Also remember Octane does have it's biased kernels....which would be similar to a final gather solution in a regular render, and they are much faster.

Knowing how to tune an Octane render makes a huge difference in time savings. Most people I see just use the defaults.
Coherent ratio for example can speed up the render by 40%....but flickers in animations. Also setting spec gloss and diffuse bounces can drastically speed things up. By default most are set to 24 bounces which can be overkill.
Thanks.
Of course we can not render this scene in PM but what we can do is render with QMC or PM solution to shows differences. Good Photon Mapping engine is 10-20 times faster then Path Tracing with similar effect but visible less quality. QMC around 5 times faster with very slight differences. Actually most PT engines use MTL (Metropolis) system and this is kind of successor of pure PT which will be also in K3.
In K2.x this scene should be slower because Kray has standard Path Tracing solution (not MTL) but the biggest problem is how to COMPARE!
I tried find on web how to compare GPU vs CPU speed in renders 3D but this is mission impossible because these are completely systems of renders. People just wasting time talking about which quality is better from CPU or GPU which is pointless.
Personally I think one GPU Titanium Z is equal to Xeon 20 Threats but I can be wrong.
Here, also nice graph I found:
Graph1.jpg
Path Tracing solution is very "sensitive" for amount of photon bounces and complexity. Please notice that fact, most people shows very fast speed studio renders from Maxwell or Octane (or basic scenes) forgetting fact that in these scenes is nothing to calculate! Scene must be complex (best choice is interior) with 10 bounce lights with many lights and full surfacing solution...then you can compare...
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Re: Octane quick test-drive

Post by thomas »

ideart wrote:GTX 760 kills all the joy working with Octane.
I'm probably running on the worst hardware on this forum, I'm starting to realise: 1150 CUDA cores on the GTX 760, and only 880 Cinebench score on my i7 4790.

And I call this PC my RenderBeast! :D
Fabian-Eshloraque
Posts: 74
Joined: Sun Nov 17, 2013 3:29 pm

Re: Octane quick test-drive

Post by Fabian-Eshloraque »

Hi,

We have been testing a lot of render engines, cpu and gpu based. These are our conclusions:

1. For 3D:

GPU is fast and good to use for viewport rendering. You dont need to buy an expensive graphics card to get results. But for final render its different. Stills okay, but when it comes to animation things are totally different. You can buy a breakoutbox full of GPU's but its very expensive and power consuming. I also noticed that pc when using GPU often crash to bluescreen. And like Janusz says, the noise stays.

Budget wise: 1 good GPU card costs 5000 euro, while a c7000 enclosure WITH 12 blades dual 6 core xeons cost almost the same.

The best type of scene is a character with fur. GPU can not handle it. Noise levels are too high compared to the fur. For example renderman has 2 renderengines to takkle this: ris and reyes. One is good for fur, the other for massive GI scenes. I know of some studios who tried GPU octane and fur but had to go back to CPU due to too high levels of noise.

2. For compositing:

Things are different here. GPU is the best. Autodesk flame uses GPU and it reaches almost realtime for the complete shot you are working on! And not the frame you are working on. And they don not use CUDA because its a layer of programming made for colorgrading and editing software.

3. Shaders:

I guess that it will come in kray. I'm not worried about it. I would just like to have micro poly displacement, like for snow etc.

Best!
mdharrington
Posts: 33
Joined: Sun Sep 20, 2009 8:21 am

Re: Octane quick test-drive

Post by mdharrington »

1 good GPU cost $5000 euro???

Quadro's are not a necessity for GPU rendering...not at all

You will see by these benchmarks...a sub $650 GPU will destroy the most expensive Quadro made
https://render.otoy.com/octanebench/res ... ingleGPU=1

Quadros only offer advantage in openGL....in rendering they are much worse
User avatar
Janusz Biela
Posts: 3265
Joined: Mon Mar 13, 2006 10:39 am
Location: Finland
Contact:

Re: Octane quick test-drive

Post by Janusz Biela »

Here my test in PT (i7 CPU, 70 min):
Vensterbank.jpg
Is longer because pure Path Tracing (no MTL). Also I do not know how to compare i7 to GPU...so maybe is fast.

QMC test 40 min:
Vensterbank_QMC.jpg
You can notice, there is very small differences (impossible to see by amateurs).
to make same noise render time will be 20 min so around 7 times faster then PT.
Differences in brightens comes from photon mapping system (this is normal) but has nothing to do with sampling.
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Re: Octane quick test-drive

Post by thomas »

Hi Janusz,

Thanks for the test! Beautiful scene by the way :D

I'm curious about the difference in brightness, due to the photon mapping system. I'm shocked to see the difference in contrast, with the same tonemapping, just by choosing a different render method! Why is this normal, as you say?

By the way, did you do both render with Kray 2? The PT render has so much more contrast and life in it!
thomas
Posts: 348
Joined: Tue Aug 31, 2010 9:59 am

Re: Octane quick test-drive

Post by thomas »

Thanks for the info, Fabian, by the way. Good to know.
Fabian-Eshloraque
Posts: 74
Joined: Sun Nov 17, 2013 3:29 pm

Re: Octane quick test-drive

Post by Fabian-Eshloraque »

Yes mdharrington,

There are a lot cheaper GPU cards, but if we need to have GPU card, we get a quadro K6000 or M6000 which rate at 5K. And yes you can buy lower cards for octane, but they are useless for other software like some compositing software or 3D render engines. (joy)

Which leads to the fact that there are 2 groups of GPU cards (or 3 because of the Titans and/or Intel Phi) and is giving me a hard time to manage our workstations. And for some Linux installations I need to trash the Cuda software layer and replace it with another. (because its not stable enough)

My guess is that the hardware guys are making a mess out of it. Types of GPU, type of software drivers, application software to run on it...
User avatar
Janusz Biela
Posts: 3265
Joined: Mon Mar 13, 2006 10:39 am
Location: Finland
Contact:

Re: Octane quick test-drive

Post by Janusz Biela »

thomas wrote:Hi Janusz,

Thanks for the test! Beautiful scene by the way :D

I'm curious about the difference in brightness, due to the photon mapping system. I'm shocked to see the difference in contrast, with the same tonemapping, just by choosing a different render method! Why is this normal, as you say?

By the way, did you do both render with Kray 2? The PT render has so much more contrast and life in it!
Both with K2. Differences in Photon Mapping (in brightens) comes form render engine and system of transporting photons . We must understand Photon mapping is interpolated process and all information is taken from averaged data. Each render engine (V-ray, Modo, etc) has different system of transport and calculations. Of course Global Illumination in all render engines (for example Maxwell, Octane, Corona, Arnold, V-ray, etc) are based on this:
https://en.wikipedia.org/wiki/Global_illumination

Some engines sucks in Photon Mapping (sorry Newtek) some are good but mathematical solution is the same everywhere.
Of course final effect strongly depends also from surfaces (especially from shaders) and post process. So all companies which sell render engine should show always pure RAW renders not gallery from artists.
We start soon totally new K3 web page and we took decision with. G to show always pure RAW renders on the front page. Standard gallery will contain all renders.
Soon K3 render :wink:
Locked