PDA

View Full Version : The ultimate Render Engine Discussion



speismonqui
09-11-2010, 04:59 PM
Hi, I've been reding a lot about render engines, however there's somethings unclear to me, I thought you guys could help me out...

It's not my intention to start a boring Software "X" vs Software "Z" battle, just to get clear and learn from the experts.

I want to talk about Vray, MentalRay, Lightwave, FPrime and Kray and the softwares involved with this render engines.

For what I (think) i know:

1. Vray and MentalRay are equal in quality, but VRay is faster.

2. If this is true, Lightwave (FPrime included) is even faster than Vray.

3. Vray and Kray offers almost the same functions. VRay uses both Max's and its own lights, cameras and materials. Kray uses Lightwave's, no need to have their own.

4. Lightwave's renderer (FPrime included) doesn't have Photons, tonemapping, instancing, gamma and linear workflow (to name a few) and other features Kray, VRay and MentalRay has.

5. For animation, I've read that Vray has some flickering issues depending on settings, MentalRay (in Softimage) has serious motion Blur issues, don't really know about Maya. Is Lightwave native renderer better for animation? How about FPrime? Kray?

6. In your expert experience, is VRay more "profesional" than Lightwave?

Thats it for now, I hope to start an interesting discussion about this RENDERERS. Maybe later we can make another one about Modo, Lux, Octane, Maxwell, Arnold, etc, etc, etc...

TNX

speismonqui
09-11-2010, 05:02 PM
oh, I forgot...
7. Architectural Interior lighting: which one do you think gives the best quality/render times?

Dreamcube017
09-11-2010, 05:30 PM
Hmm... this is interesting. Coudl someone answer this.

silviotoledo
09-11-2010, 06:44 PM
I don't have all the answers, but I have some:


As NATIVE render engine Lightwave have the best render and now VPR is kinda an FPRIME.

it seems FPRIME replicates LW render in a faster way i really don't understand how.

MR is faster than VRAY

VRAY RT is going faster than MR, but still in progress...

KRAY gives almost the same quality as V RAY, sometimes better, but V RAY is more developed.

About Photons, I think all they have, but it seems POST processing in V RAY and MR is more advanced than in LW, that's why they get better results. And they also have better shaders.

V RAY is the best for Architectural, but see that K ray does also a great job. MR is also too good.

For me ( and the comunity will kill me ) V RAY and MR have better rendering quality than Lightwave and I'd love to see V RAY and MR for lightwave some day.


Physically correct Photorreal ( Maxwell, Fryrender, Arnold... ) takes too much time to render.

No perfect renderer is available, so post is required :)

Also consider that people have different styles and preffer different renderers. So too hard to say wich one is better.

Dreamcube017
09-11-2010, 08:36 PM
Lt's not forget Octane and some other GPU renderers being developed. As far as I'm concerned, ANY GPU renderer is better than a CPU only renderer just because it's a bit quicker... but that's mainly because I'm impatient.

sampei
09-11-2010, 09:38 PM
isn't linear workflow something you do in post, if so what's the point of having it in a renderer ? and vray and mental ray have better shaders than all the pretty LW nodes ? in what ways are they better ? guess this is one discussion that keeps popping up every now and then.

Dreamcube017
09-12-2010, 12:55 AM
I wish I could afford FPrime... but if I'm going to get that, I might as well just get LW10.

Captain Obvious
09-12-2010, 03:47 AM
1. Vray and MentalRay are equal in quality, but VRay is faster.
Compared to... what, each other? Doing what? If you want a straight-forward and quick set-up, Vray is a better choice. If you need bespoke shaders and whatnot, then mental ray is probably a better choice.




2. If this is true, Lightwave (FPrime included) is even faster than Vray.
Lightwave is usually not faster than Vray, but it can be pretty fast. Lightwave has the same advantage mental ray does, though: you can customize it more. The node-based stuff can let you do some crazy stuff.




4. Lightwave's renderer (FPrime included) doesn't have Photons, tonemapping, instancing, gamma and linear workflow (to name a few) and other features Kray, VRay and MentalRay has.
Lightwave does have photons. That's how it does caustics. And it's got some of the most advanced tone mapping available; you just have to do it yourself... With Denis Pontonnier's node pixel and image filter (http://dpont.pagesperso-orange.fr/plugins/nodes/DP_Filter.html#DPFilter) you can create pretty much arbitrary tone mappers. Save the nodes to a file and then it's easy to add them to any project. You can actually do multi-layer compositing and grading inside the render engine with Lightwave. Not that you'd want to, but the capability is there. You can do some really advanced stuff.

Cageman
09-12-2010, 04:03 AM
For me ( and the comunity will kill me ) V RAY and MR have better rendering quality than Lightwave and I'd love to see V RAY and MR for lightwave some day.

It really boils down to what you need to do given the time you have. We are currently wrapping up a 10 minute 720p 100% CG-movie that was rendered within 2 weeks using LWs renderengine (including wrangling). For what we needed to do and the timeconstraints we had, we would not have gotten the time/quality ratio with any other engine. Fact is that we have now almost reached 1 million rendered frames (many different passes + renderbuffers etc) and LWs renderengine has been incredibly stable and reliable. We render with GI btw...

As a comparsion, when we did another movie/trailer thingy completely in Maya/Mentalray, it was more than two weeks to render for just about 1 minute of full CG (including wrangling).

So, it all depends on the project, really.

Elmar Moelzer
09-12-2010, 05:49 AM
As far as I'm concerned, ANY GPU renderer is better than a CPU only renderer just because it's a bit quicker... but that's mainly because I'm impatient.
As someone who does have experience with actually developing GPU renderers, I can only tell you that this is not true.
GPU renderers are awesome when speed and only speed is needed and then only when the scene does not surpass a certain level of complexity.
The more complexity increases the harder you have to work arround the limitations of the GPU and the closer you get towards the point of diminishing returns.
An example out of the volumetric rendering area:
With a software renderer it is very easy to combine geometry and multiple volumetric objects at will. They can intersect each other, cast correct shadows onto each other and so on.
In a GPU renderer situation the whole thing is much more complex and even specialized university departments are still having a hard time getting results equal to a CPU renderer here and if they do, the whole process is very involved and quite complex (and requires the use of CUDA, e.g., which brings other issues).
To much complexity is a bad thing and things will get slower the more complexity you add. The whole architecture also becomes less flexible that way and it becomes harder to add new features, or for 3rd parties to add their own features.
All this leads again to the point of diminishing returns. So to go back to LW, once the scene complexity reaches a certain point, LWs software renderer will surpass the GPU renderer in pretty much every aspect, including workflow.

If I were to bring GPU rendering to LW, I would use it for preview only. So to preview the general appearance of the whole scene (with limitations that make it unusable for final shots), or to preview single, extracted elements of the scene at a close to full quality (e.g. just the volumetrics, or just the geometry, or just a single surface, etc). So basically a GPU based VIPER would be the best way to use the GPU right now, IMHO.

Dreamcube017
09-12-2010, 07:48 AM
Hm, a preview render would be nice though for doing shaders and lighting. It's in LW10 though so that's cool.

Dreamcube017
09-12-2010, 08:09 AM
I took a look at Kray, what's so different about it compared to the Lightwave default renderer anyway?

speismonqui
09-12-2010, 01:56 PM
wow, thanks for the replies.

I really thought both Mental Ray and Vray were slower tha Lightwave. I've used them all in small projects and stills and the quality/render times were one of the main reasons for me to switch to Lightwave. Maybe I just blindly love lightwave!

Also, I've read a lot about Mental Ray in Maya is very slow. A friend of mien (maya user) hates it, and wishes to have the LW renderer in maya :D

I think we can all agree that "A project" determines which software/renderer is best suitable for a specific production. Its really not a matter of which one is better than other.

:thumbsup:

Dreamcube017
09-12-2010, 02:19 PM
Yeah, from the little I've done, Lightwave renders GI and things pretty well and quickly too... for the most part anyhow. *sigh* I still wish I could afford FPrime or LW10 though... oh well, I'll just be patient and maybe I'll get it oneday.

silviotoledo
09-13-2010, 08:54 AM
yeah! all depends on the projects, time and money.

Lightwave is good, so easy to use, fast and cheap.

It's not the best " photorreal renderer " but the others does not have this combination.

Maybe some post processing nodes will increase lightwave rendering results. I guess they started working on this on CORE.

Actually V ray, K ray, Mental Ray, Final Render and others have something is post ( or maybe tone mapping ) that lightwave doesn't have, as default, that gives a better " photorreal " look.

Infinite
09-13-2010, 09:03 AM
You forgot to mention Arnold. Newtek should get in contact with these guys.

The render engine is incredible. It stomps all over MR and Vray in terms of speed and quality.

A bridge into LW or CORE would be a very clever thing to do....

Dreamcube017
09-13-2010, 09:17 AM
Hm that Arnold renderer is interesting. Never heard of it. I never even knew Sony made a renderer... I can't seem to find the actual site though. I just find other sites talking about it. Hmmm.

silviotoledo
09-13-2010, 09:26 AM
Arnold is in development by SONY, so it seems it will be an inhouse tool only.

In the past, Arnold have been included in Project Messiah wich started as an animation plugin for lightwave and now is a complete animation solution.

In comparison to Mental Ray, it seems Arnold is around 4 times faster!

But V ray for maya is comming with V Ray RT

See tests here:

http://www.cgauiwtalk.com/showthread.php?14797-Arnold-Render&p=75690

Dreamcube017
09-13-2010, 09:33 AM
Yeah I saw that thread.

Inhouose... BOOOOOOO!!! ... Oh well.

Hmm Project Messiah? Never heard of that. I'll look it up.

VRay RT... I remember watching a video on that sometime ago... maybe about a year or so.

So far, the best I'm hearing about as far as LW is concerned is FPrime though although VRay RT does sound promising as far as renders in general go.

Oh, you're forgetting all the other tiny renderers like Aquis or it's called something like that.
MakeHuman uses it along with one other.

Lightwolf
09-13-2010, 09:37 AM
Arnold is in development by SONY, so it seems it will be an inhouse tool only.
Actually, Arnold is licensed by Sony, but developed by solid angle (which is the company of Marcos Fajardo).
http://solidangle.com/
And yes, it can be purchased, it's just not really advertised. I guess it's targeted at larger studios only at the moment.

Cheers,
Mike

Dreamcube017
09-13-2010, 09:47 AM
sooo... this is KIND of in the same area... but anyone heard of MachStudio Pro? It's like an animation thing that uses DirectX 11 to render things.

Now I know DX is no MR or VRay or whatever, but I still think it's valid... DX is a way or rendering things... so is OpenGL... and OpenCL... and CUDA for that matter.

Don't the realtime GPU solutions get any love?

Elmar Moelzer
09-13-2010, 12:29 PM
GPU does not necessarily equal realtime.

speismonqui
09-13-2010, 01:51 PM
i still remember what i felt the first time i saw this, my jaw still hurts

http://www.youtube.com/watch?v=VNyknZ2zrsM&feature=related

this was about 5 or 6 years ago.

jrandom
09-13-2010, 01:57 PM
I took a look at Kray, what's so different about it compared to the Lightwave default renderer anyway?

KRay is amazing at handling a lot of blurry reflections (to the point that you can ditch specular highlights entirely). It tends to fall down on anti-aliasing, though. It also has a blazingly-fast non-interpolated Monte Carlo engine (compared to Lightwave's) -- good if you're getting too much splotchyness in an Interpolated or Final Gather render.

I'm so curious as to what's causing the different speedups and slowdowns that I've ordered the book "Physically Based Rendering - Theory and Implementation" and will try to write my own renderer, just to learn. I'm curious to see how fast I can make it and how much physical accuracy can be sacrificed and still get a good render.

I don't expect I'll be able to compete with the pro renderer packages out there, but what total-geek programmer hasn't wanted to write their own renderer at some point or other? :P

speismonqui
09-13-2010, 03:06 PM
keep it up then, wish you luck and please... make it for LW not for AD :p

jrandom
09-13-2010, 03:17 PM
keep it up then, wish you luck and please... make it for LW not for AD :p

Well, of course. I don't own any AD software and since I no longer have to pirate software (gots me a real job!), I don't have any AD software on my computers either. :)

It will, however, most likely be 64-bit-only, which means targeting Lightwave 10.

geothefaust
09-13-2010, 03:53 PM
Hmm Project Messiah? Never heard of that. I'll look it up.




I use Messiah for all my rendering needs these days.

Dreamcube017
09-13-2010, 04:14 PM
Hm I took a look at the site. It looks pretty cool. Is it easier for doing animation or something... or does it have a better renderer fr doing animation or... a sky and nature system or something... or... what?

I saw the videos, but from what I saw, it wasn't that epic to me... just my opinion.

speismonqui
09-13-2010, 04:19 PM
Taron also uses Messiah for rendering, eventough he's a LW user.

http://www.taron.de/LilDevil.htm

interesting link

monovich
09-13-2010, 09:13 PM
Lt's not forget Octane and some other GPU renderers being developed. As far as I'm concerned, ANY GPU renderer is better than a CPU only renderer just because it's a bit quicker... but that's mainly because I'm impatient.

As long as you don't have to set up you shaders all over again. Not to mention animation support, etc.

I haven't re looked lately but I haven't seen a production ready gpu renderer.

Elmar Moelzer
09-14-2010, 03:57 AM
It will, however, most likely be 64-bit-only, which means targeting Lightwave 10.
Why would a 64 bit version require LW10? There is a 64 bit version of LW9.6 too...
Also, if you are doing a GPU renderer, then you might not even hit the memory limits of 32bit before running out of memory on the graphics card.


Taron also uses Messiah for rendering, eventough he's a LW user.
Thats in part because he is, or at least used to be, the product manager for Messiah.
I think he is using LW for rendering as well, at least occasionally (he used to anyway).

geo_n
09-14-2010, 05:31 AM
3dmax new features.
Iray cpu AND gpu enhance renderer now built in for subscription users.
http://area.autodesk.com/blogs/ken/3ds_max_2011_subscription_advantage_pack_revealed_ iray_physx_substances

Kray 2.2 soon to be released - whats cooking in this release?
http://www.kraytracing.com/joomla/phpbb/download/file.php?id=3383&mode=view

Octane renderer beta 2_4 - fast development
http://www.refractivesoftware.com/index.php


modo renderer = sexy and easy
http://www.youtube.com/watch?v=qhi4VMR1xrI

vray render - ultimate renderer
http://vimeo.com/7809605

Elmar Moelzer
09-14-2010, 08:39 AM
Iray is so not impressive at all. First, it seems very slow (30 minutes for this?), GPU or not.
Also, with all the money that Autodesk has, one should expect that they would be able to find someone with better English to present the renderer.
I think that LWs renderer beats Modos in about every aspect, but maybe that is just me.
Vray is good, but I not as easy to use as LWs renderer.

jrandom
09-14-2010, 09:03 AM
Why would a 64 bit version require LW10? There is a 64 bit version of LW9.6 too...

Oh, right! Sorry. Mac user here. Keep forgetting Windows exists. :)



Also, if you are doing a GPU renderer, then you might not even hit the memory limits of 32bit before running out of memory on the graphics card.

If I can find spots to use the GPU, I will certainly do so, but I don't want to artificially restrict the renderer to video RAM. I can just tell that would come back to haunt me if I went down that path.

Elmar Moelzer
09-14-2010, 09:12 AM
If I can find spots to use the GPU, I will certainly do so, but I don't want to artificially restrict the renderer to video RAM. I can just tell that would come back to haunt me if I went down that path.
If you want to use the GPU one day, then make sure you plan for this right from the beginning, especially when you are writing your shaders so they can be easily moved over to the GPU. Adding GPU support as an afterthought will cause you nothing but headaches.

jrandom
09-14-2010, 09:25 AM
If you want to use the GPU one day, then make sure you plan for this right from the beginning, especially when you are writing your shaders so they can be easily moved over to the GPU. Adding GPU support as an afterthought will cause you nothing but headaches.

Absolutely. I'm going through OpenCL tutorials this very moment. :) What I don't know is just how much I'll be able to use the GPU. Hm... y'know, motion blur on deformable objects could probably benefit from that.

speismonqui
09-14-2010, 11:38 AM
I think that LWs renderer beats Modos in about every aspect, but maybe that is just me.


:agree:
so am I
do you feel the same way about Fprime vs Modo's renderer?

Elmar Moelzer
09-14-2010, 01:18 PM
do you feel the same way about Fprime vs Modo's renderer?
Fprime has a very different quality to it. I dont have so much experience with it, as I never bought a seat of it. I only ever saw it demoed by other users. I think that even with VPR, Fprime will have a place in the market and I even think that Steve Worley could bring Fprime foreward once again and give even users that have VPR an incentive to buy Fprime (or to upgrade their copy).
There are still a few holes that Fprime can fill.

Captain Obvious
09-14-2010, 01:50 PM
I think that LWs renderer beats Modos in about every aspect, but maybe that is just me.
It's just you. Lightwave's renderer certainly has a few strengths compared to modo. But for things like texture filtering and anti-aliasing, modo kicks some serious behind. I am rather infatuated with the fact that you can turn up the AA to, say, 128 samples without significantly impacting performance. I've never seen a ray tracer with as good anti-aliasing as modo. But I think Lightwave's GI is generally better, and the nodes are pretty sweet.

Edit: and in modo, texture filtering works through recursions. In LW, as soon as the ray has passed through a single recursion, texture filtering is basically turned off. In modo, this is not the case. This means you get nice and clean texture filtering even through transparency and reflections.

HypeTsu
09-14-2010, 09:48 PM
sooo... this is KIND of in the same area... but anyone heard of MachStudio Pro? It's like an animation thing that uses DirectX 11 to render things.

Now I know DX is no MR or VRay or whatever, but I still think it's valid... DX is a way or rendering things... so is OpenGL... and OpenCL... and CUDA for that matter.

Don't the realtime GPU solutions get any love?

MachStudio Pro is much more than a rendering engine.
It is a complete "finalling" suite placed in between animation to compositing.

You can place and manipulate your lights, build and adjust your shaders, control your camera and render out any combination of passes to - almost - any format.

Further, in the new MachStudio Pro 2.0 (currently in beta and revealed at Siggraph few weeks back) there exists support for Renderman and Mental-Ray in such a way as to allow for an image constructed from both; CPU rendering (MR + PRMan) and GPU rendering (MSP) to be put together so that:

At any given time and dependent on desired look, one can decide which portion will be rendered on the GPU and which on the CPU.

So for instance; only the wine glass can be ray traced by the CPU using MR; while the rest of the room can be rendered by the GPU using MSP. Or another example; render all the passes on the GPU using MSP, while getting the GI pass from MR (final gather) using the CPU. Or render the car using MSP, while getting a ray traced reflection pass out of Renderman.

If you want, you can even do the entire final render in your CPU rendering engine of choice, the point is that throughout your "finalling" work, you enjoy a totally interactive workflow and a view port that gives you 90% of your final image quality instantly.

So yes; GPU tools deserve more love :-)

Dreamcube017
09-14-2010, 09:57 PM
Wow that's pretty awesome.

HypeTsu
09-14-2010, 10:01 PM
GPU does not necessarily equal realtime.

True.

GPU doesn't mean real time.

Let's put order in the term Real Time.

We can start by saying " The Only Real Time is; NOW" :-)

Seriously now...

In games: real time is 60 frames per second
In terms of movie/video playback: real time is 24-32 frames per second
Generally, anything bellow 15 frames per second is termed "near real time" or "Interactive".

But remember that after all, we talk of real time terms of responsive UI and viewport, and for that; real time is about 5 frames per second = smooth moves and instant render gratification.

HypeTsu
09-14-2010, 10:14 PM
I haven't re looked lately but I haven't seen a production ready gpu renderer.

MachStudio Pro is a production ready rendering engine GPU.

See these two examples: Zoic Studios using MachStudio on NBC's "V" (see here: http://blog.digitalcontentproducer.com/briefingroom/2010/03/29/v-the-visitors-and-zoic-studios-join-forces/ and Delacave Paris using MachStudio on "Cinderrella: The Gal from Out West" (see here: http://features.cgsociety.org/story_custom.php?story_id=5679

One is a VFX example and the other is Animation example...

Elmar Moelzer
09-15-2010, 05:39 AM
To me realtime is final output quality at interactive framerates.
Ideally those framerates would be at the desired output framerate (so e.g. 24 fps for film).
Games are not always 60fps, btw.

Tartiflette
09-15-2010, 10:06 AM
MachStudio Pro is a production ready rendering engine GPU.
No Mac version unfortunately, so it's a no go for me... beside its price, really expensive. :(


Cheers,
Laurent aka Tartiflette :)

Dreamcube017
09-15-2010, 07:09 PM
Wait, the Mac version of Machstudio Pro isn't as good as PC's?... this may be because it's not using DX11, it's using OpenGL. Maybe when OpenCL or something is finished it'll be better.

I agree, MachStudio Pro is kind of hefty in price but I'd really like to try it.

They have a demo, but my card doesn't have a gig of VRAM, it has like 800.

Tartiflette
09-16-2010, 03:50 AM
Wait, the Mac version of Machstudio Pro isn't as good as PC's?... this may be because it's not using DX11, it's using OpenGL. Maybe when OpenCL or something is finished it'll be better.
No, there is just NO Mac version of Mach Studio Pro. ;)

See here... (http://www.studiogpu.com/forums/viewthread/5/)

So if you want to run Mach Studio Pro, you have to run Windows and that's all. And i don't think it has anything to do with OpenCL or not as they are not using Cuda for the render engine but DirectX 11, which is a totally different beast, kind of competitor to OpenGL.


Cheers,
Laurent aka Tartiflette :)

OFF
09-16-2010, 07:48 AM
RenderGin – Beta
http://www.youtube.com/watch?v=F8pg7EtkXKw

http://www.numenus.de/download.html

RenderGin native fileformat (*.rg)
Wavefront Object (*.obj)
Standford Polygon Format (*.ply)
Autodesk FBX (*.fbx)
Autodesk 3D Studio Max (*.3ds)
Collada 1.4 (*.dae)
Initial Graphics Exchange Specification (*.iges / *.igs)
Rhino 3D Model (*.3dm)
OpenSceneGraph Ascii (*.osg)
OpenScenegraph Binary (*.ive)
Design Workshop Database (*.dw)
Lightwave Object (*.lwo)
Standard for Exchange of Product Model Data (*.stp/ *.step)
Virtual Reality Modeling Language (*.wrl)

Elmar Moelzer
09-16-2010, 10:26 AM
I tried installing it, but the installer cancelled. Core2Duo 2Ghz, Geforce GTS 8800M, 3GBs of Ram.

Ok, it worked after I made a complete install.
What a complete waste of time!
VPR is much faster.
OFF, have you actually tried that thing at all?

OFF
09-16-2010, 10:39 AM
you tried x32 or x64 version?
for me It was installed perfectly, rendering 120mb NY City object with mad speed

Elmar Moelzer
09-16-2010, 10:43 AM
I tried the 32 bit version.
As I said, the installer worked after I tried again with the "complete" option istead of custom.
I used one of our VoluMedic generated Meshes at full quality with it (one million plus polys). I have used the same to test VPR. VPR was faster and looked much better.

Dreamcube017
09-16-2010, 04:38 PM
Well I guess that sums it up... I have to test this though. Hmm...

speismonqui
09-16-2010, 05:10 PM
I tried the 32 bit version.
As I said, the installer worked after I tried again with the "complete" option istead of custom.
I used one of our VoluMedic generated Meshes at full quality with it (one million plus polys). I have used the same to test VPR. VPR was faster and looked much better.

can you post some images?

jrandom
09-16-2010, 05:10 PM
Well kids, I just got this book (http://www.amazon.com/gp/product/0123750792/ref=pd_lpo_k2_dp_sr_1?pf_rd_p=486539851&pf_rd_s=lpo-top-stripe-1&pf_rd_t=201&pf_rd_i=012553180X&pf_rd_m=ATVPDKIKX0DER&pf_rd_r=0YRVK9VWQKBFFE6A34VQ) in the mail.

Wish me luck. :)

silviotoledo
09-16-2010, 05:27 PM
Lots of physically correct renderers going realtime

KEYSHOT
http://www.keyshot.com/


ARION ( Fryrender )
http://www.youtube.com/watch?v=V-aKiNMzf24
http://www.youtube.com/watch?v=fIOHgk2Jut4


VRAY REALTIME
http://www.youtube.com/watch?v=8kXDiB1cdDg
http://www.youtube.com/watch?v=eRoSFNRQETg


OCTANE RENDER
http://www.youtube.com/watch?v=xHqRLLbfQt0

Cageman
09-16-2010, 05:50 PM
None of those are realtime though, since the quality of shading is drasticly reduced everytime something changes. Close to realtime or Interactive is a much better description of what actually goes on.

Lightwolf
09-16-2010, 05:55 PM
None of those are realtime though, since the quality of shading is drasticly reduced everytime something changes. Close to realtime or Interactive is a much better description of what actually goes on.
I like the term progressive a lot...

Cheers,
Mike

silviotoledo
09-16-2010, 06:35 PM
yeah, the correct term is PROGRESSIVE wich means realtime feedback :) and fast rendering.

Cageman
09-16-2010, 07:17 PM
yeah, the correct term is PROGRESSIVE wich means realtime feedback :) and fast rendering.

Fast, yes, but realtime, no. I do have Octane and I can easily create content that a game-engine will render in 60fps, but Octane might need several seconds or even minutes to refine.

Sensei
09-16-2010, 07:27 PM
Realtimes really depends on complexity of scene, and forever will be.
Scene that has 1000 polygons will be realtime, but with 1 mln or 10 mln, on CURRENT machines will not be realtime.

After 1,2,3 years it might be realtime.

But then, you will complain that scene with 100 mln polygons is not realtime.. and we're back in home.. ;)

Dreamcube017
09-17-2010, 01:19 AM
Fun fact... did you konw that the game Uncharted 2 for the PlayStatino 3 has over a million polygons on screen at once and it doesn't go below 30 FPS?

I think we're a bit closer to high poly realtime models and things than you may think.

I'm pretty sure that with a good GFX card, you can get some pretty decent poly models (40,000+ per character) in a scene without the game lagging too much.

Although what makes things look so good is more texturing and lighting than actual polycount.

So shaders and lighting play a large part.

Elmar Moelzer
09-17-2010, 05:34 AM
First, dont believe everything marketing tells you.
Second, in order to render complex scenes, even modern 3d engines require a lot of manual labour by the developer and often many versions of the same object for LOD in a distance. Shaders have to be highly optimized and sacrifices in quality have to be made.
In addition to this, even modern engines still preprocess a lot.
Dont expect to take your LW- scene into the engine of Uncharted and get 30FPS. This is not how it works.

Most of the so called "realtime" renderers nowadays are actually progressive renderers, as has been mentioned. I personally dont like the term realtime very much, because it is a very stretchable term.
VPR and Fprime are also progressive renderers so they are also "realtime" as much as the renderengines listed here.

The other thing that I absolutely dislike is the fact that people assume that whenever they hear "GPU accelerated", they think that this automatically means that they will get the same feedback and rendertimes for a fully rendered frame as they would in a 3d game engine.
This is almost never the case (I say almost, because there are examples, but they do have their limitations, also).

Dreamcube017
09-17-2010, 06:44 AM
I konw I can't just pull any random high poly object into an engine. I've worked with engines before and I know how they work and I'm aware that things are done differently. (shaders for example are done different as there's really no such thing as a true procedural shader... there are things that come close, but not like in nonrealtime applications.

I know games aren't quite at a nonrealtime 3D quality yet, but I till think it's improving pretty well with some nice results.

Elmar Moelzer
09-17-2010, 08:43 AM
but I till think it's improving pretty well with some nice results.

True, but at the same time the demand on the rendering quality is increasing as well.

Captain Obvious
09-17-2010, 09:09 AM
Realtimes really depends on complexity of scene, and forever will be.
Scene that has 1000 polygons will be realtime, but with 1 mln or 10 mln, on CURRENT machines will not be realtime.
Weeeell, keep in mind that scene complexity is not the same thing as polygon count. Well-optimised ray tracing is not heavily affected by the number of polygons. There are some special (CAD-industry) solutions that can ray trace hundreds of millions of polygons in near-realtime. Probably using clusters / thin clients, but still. I read about some software that Boeing use to view their whole master CAD files, and basically they couldn't build a rasteriser fast enough. A whole Boeing 777 is probably a couple of hundred million triangles when you have every nut and bolt and wire, and they needed something capable of viewing it in interactive rates.



Fun fact... did you konw that the game Uncharted 2 for the PlayStatino 3 has over a million polygons on screen at once and it doesn't go below 30 FPS?
Fun fact. Did you know that some of the high-res assets used for a game I used to work on weighed in at 45 million triangles? And that it still ran at about 30 fps in modo, using a Quadro FX 4500... :)




Also, regarding the whole "GPU performance" and real-time stuff. Imagine a heavy high-res Maxwell scene. This one thing we rendered took about 30 hours across 26 eight-core machines, at a resolution of about 5k by 3k. If we want to render that in real-time at 20 frames per second at a resolution of 1k, in a GPU-accelerated renderer. Then we'd need a GPU that's more than two million times faster than an eight-core Xeon.

Elmar Moelzer
09-17-2010, 09:35 AM
Probably using clusters / thin clients, but still.

Yes, the university in Vienna is/was actually a leading developer of "realtime" raytracing solutions for this very purpose.
I still have the paper here somewhere, but it indeed does rely on clusters of CPUs working in tandem and it is meant for basic visualization, not photoreal high end movie rendering.

Sensei
09-17-2010, 09:48 AM
Weeeell, keep in mind that scene complexity is not the same thing as polygon count. Well-optimised ray tracing is not heavily affected by the number of polygons.

For real-time/progressive renderer polygon count is essential, because they have to build kd-tree after every serious change in scene, such as moving object..
Fast ray-tracing is result of well optimized kd-tree, and building optimized kd-tree is very slow. There is very little what you can do inside of ray-tracing routines! Flying through kd-tree is just one subtract and one divide per node and can be written in 10 lines of c/c++ source code or less..
Off-line renderers like F9/F10 LW renderer or Kray, don't care much, 1-5 second delay after pressing button, spend on building kd-tree, is nothing for them if whole rendering will take f.e. >15 minutes..

Do you know how long takes scanning polygons in LightWave?
507408 polygons LWSDK scanPoly() call-back, doing nothing else than asking LW to return all polygon data including pntOtherNormal() on Core i7 930 takes 0.312 second.. (checked minute ago, in TrueHair Preview)
JUST reading geometry from LW.
What can renderer do with it? Reverse engineer LWSDK data structures and assume that polygon is in continuous array in memory, completely ignoring official LWSDK functions?

Any movement, transformation of such object, will be resulting in at least 0.312 second delay for interactive renderer..

And after scanning polygons, we need to build kd-tree, that's another delay before previewer will even start rendering anything..

Elmar Moelzer
09-17-2010, 10:01 AM
Yet VPR is rather nice and quick, at least on my machine. I think there are still a few things that can be done to make this faster.
Fprime also does rather well in regards to speed.
There are a few things I have been thinking about, but I dont have the time, nor the energy to look into it right now.

Captain Obvious
09-17-2010, 10:10 AM
What can renderer do with it? Reverse engineer LWSDK data structures and assume that polygon is in continuous array in memory, completely ignoring official LWSDK functions?
That's probably what FPrime does.

Lightwolf
09-17-2010, 10:50 AM
That's probably what FPrime does.
Highly unlikely, are there's still a delay when you start FPrime.

I'd assume that it tracks position changes differently from deformations and it may also not even using a kd-tree (which isn't the end of all acceleration structures anyhow, interestingly enough Arnold went back to a nested bounding volumes hierarchy after using kd-tree prior to that).

Cheers,
Mike

Sensei
09-17-2010, 11:41 AM
Highly unlikely, are there's still a delay when you start FPrime.


Unlike TrueHair Preview, Fprime builds kd-tree on Master Layout thread, so it has to run as hell, to not affect interactivity. I had this way at the beginning, but later moved it to rendering threads, to not slow down interactivity..

Hacking scanning object data, could reduce just these 0.312 seconds from mine example, but still kd-tree building time is there slowing down things..

Elmar Moelzer
09-17-2010, 12:32 PM
it may also not even using a kd-tree (which isn't the end of all acceleration structures anyhow
True there are other structures that may be more suitable for certain situations.

erikals
09-17-2010, 03:14 PM
btw, does 3Dmovie renders for cinema require 2 separate renders?
if so it will increase rendertime.

also UltraHD is around the corner, in 3D... :O

Elmar Moelzer
09-17-2010, 03:26 PM
btw, does 3Dmovie renders for cinema require 2 separate renders?

Yes, you render one sequence for the left eye and one for the right eye.
They are then projected by two projectors that are using polarisation filters (as are the 3D glasses). I think there is now a new method that only requires one projector, but for that two, you would have to render two separate sequences.

beverins
09-19-2010, 10:25 AM
drifting from the realtime discussion, but noticed on the first post a query about "linear color space" - LW has LCS via node control. A demo is posted in HDRI magazine and a good demo it is too.