PDA

View Full Version : A NEW fR4 GPU Video by cebas


MKabesh
11-08-2012, 04:54 PM
Check out fR 4GPU's NEW Video!

http://www.cebas.org/blog/2012/11/07/active-mechanical-engineering/

Stay tuned for more :)

cebas Team

TwinSnakes
11-08-2012, 06:03 PM
Is there any GPU render engine that does not perform well on a 4 GPU system?

Is cebas really THAT far behind?

Edwin Braun
11-08-2012, 06:13 PM
Is there any GPU render engine that does not perform well on a 4 GPU system?

Is cebas really THAT far behind?

Depends how you define it I guess :)

We use already CUDA 5 and that was released October 15, 2012 so we are not behind on that :)

General speaking, you're right but when you look at having really useful rendering solutions on GPU that might have potential to replace the main renderer in a complex production pipeline there is nothing out there! Our system was the first to show true geometry instancing and blurry refraction/reflection effects based on a Physically accurate Micro Facet model on GPU.
There is so much more to a renderer than just "preview". We're working hard to get proper material support and so on .... so if we are behind - I would say NVIDIA is also behind :) we can only use what is exposed to us .... and a real full blown rendering system needs more features and access to the system than we have now.

-ed

DePaint
11-08-2012, 06:40 PM
Much better demo that your last one, Edwin.

I like the part where the 3D model/gears turn into glass.

Could you show some more stuff like this?

Maybe with a more complex CAD model than this one? (e.g. an internal combustion engine or gas turbine, or something detailed like that...)

Good work! Good luck...

MAN0
11-08-2012, 06:45 PM
can you show some indoor realtime rendering :thumbsup:

AbnRanger
11-08-2012, 07:40 PM
Is there any GPU render engine that does not perform well on a 4 GPU system?

Is cebas really THAT far behind?Friends don't let friends toke and type. :rolleyes:

http://gifsoup.com/imager.php?id=217269&t=o (http://gifsoup.com/view/217269/smoking.html)

FinalRender 4 GPU....is a N-A-M-E. It doesn't imply usage of 4GPU's anymore than C4D 13 means it uses 13 CPU's.

Anyway, Edwin, I'm glad to see things shaping up well on it so far.

TwinSnakes
11-08-2012, 07:49 PM
FinalRender 4 GPU....is a N-A-M-E. It doesn't imply usage of 4GPU's

hahaha..dude you got me. (LOL) I should have looked more carefully. :D

nildoe
11-09-2012, 06:26 AM
Hi Edwin,

very nice video, I got a few questions:)

1-Is you fr 4gpu renderer CUDA only or OPENCL as well?
2-Will have the same limitations as other CUDA gpu renderers like the texture limits? (for example we use Octane in our studio, and currently at least with the GTX 5xxseries and below, we get only 64 textures RGB)
3-In this video what VCard was used?
4-Does it (or will it) allow for MultiGpu setup to render an Image/animation
5-Will it support deformations (skin, etc)
6-Do you have any plans to support realtime MBlur as well?
7-What about render elements, will they be supported?

Nildo

Edwin Braun
11-09-2012, 05:15 PM
Hi Edwin,

very nice video, I got a few questions:)

1-Is you fr 4gpu renderer CUDA only or OPENCL as well?
2-Will have the same limitations as other CUDA gpu renderers like the texture limits? (for example we use Octane in our studio, and currently at least with the GTX 5xxseries and below, we get only 64 textures RGB)
3-In this video what VCard was used?
4-Does it (or will it) allow for MultiGpu setup to render an Image/animation
5-Will it support deformations (skin, etc)
6-Do you have any plans to support realtime MBlur as well?
7-What about render elements, will they be supported?

Nildo

1- Yes, for now we have a plan for OPENCL support. But for now it is CUDA only.
2- Yes, same hardware same restrictions, Kepler is 256 Textures total
3- Same as always a Quadro 5000
4- Yes we do support MULTI GPU
5- Why not ?? Yes
6- Yes
7- that's our speciality no one beats us on that !

-ed

nildoe
11-12-2012, 07:28 AM
Thx for the reply edwin :), looking really promising :).

michibe
11-12-2012, 08:16 AM
As an Octane Render and Maxwell (for special cases only) user I have my render of choice and it runs very fast, reliable and stable with regular updates but competition and innovation is always welcome. :)

CerberusC
11-12-2012, 12:49 PM
Hi Edwin.

The render engine looks awesome.
Maybe this can be the reason to change from iRay/Mental Ray to Final Render.

I have some questions:

- Do this GPU render engine suffer from the video card RAM limit like the others?
- Do you support proxys? (in the terms to optimize the RAM usage, so we can put billions of polygons in 10Mb)
- Will this be included in the standard Final Render license? (like vRay RT inside vRay)

Thanks in advance and congrats.

Cheers.

Edwin Braun
11-12-2012, 07:14 PM
Hi Edwin.
The render engine looks awesome.
Maybe this can be the reason to change from iRay/Mental Ray to Final Render.
I have some questions:
- Do this GPU render engine suffer from the video card RAM limit like the others?
- Do you support proxys? (in the terms to optimize the RAM usage, so we can put billions of polygons in 10Mb)
- Will this be included in the standard Final Render license? (like vRay RT inside vRay)
Thanks in advance and congrats.
Cheers.

Hello,

yes RAM has to be available on the GPU card - that is where the power comes form.
Your question about the proxies is a bit unclear, we do support proper instancing including instancing of proxies. This would save you memory. The whole idea about GPU is to have it all close bye - so all should fit onto the card that's the most effective way.

Yes finalRender 4 GPU will come with this GPU mode and trueHybrid™ mode. For existing users it will be a normal upgrade.

-ed

damjan
11-13-2012, 01:25 AM
@Edwin - i am sure you know about the centileo render engine - what is it with the gpu chaching they do (accessing 10 times more ram than the gpu has, trough mysterious caching algorithms)

have you tried such thing? is it possible at all?

another thing - how about corona renderer - its still alpha - but it uses basically that what we know from vray - Lightcache, and then in the primary engine, path tracing with the lightcache data - and it is very fast (and cpu only)..
wouldn't such kind of mixed approach be interesting for your new gpu system?

i am sure you work hard on this, and its just my two cents, but i dont see anything that makes your engine different from the rest - and i know your integration into max was always good - but for me it is not enough to change my pipline..

one last thing - and this is something i dont understand why other engines care so little about (except maxwell maybe - havent tried this) - the sky!
most use since 10+ years the preetham model - maybe with modifications, but still..
and they dont expose enough parameters (i read once there are about 50 variables?) anyway - mostly we get 3 and if we are lucky a horizon slider..
since years i fiddle with all the parameters and try everything possible to get some nice, not over saturated colors..
i know there are hdr etc - but having a flexible sky is imperative! ;)
..and most fail at this.. (terragen sky is very nice though)

AbnRanger
11-13-2012, 05:48 AM
@Edwin - i am sure you know about the centileo render engine - what is it with the gpu chaching they do (accessing 10 times more ram than the gpu has, trough mysterious caching algorithms)

have you tried such thing? is it possible at all?

another thing - how about corona renderer - its still alpha - but it uses basically that what we know from vray - Lightcache, and then in the primary engine, path tracing with the lightcache data - and it is very fast (and cpu only)..
wouldn't such kind of mixed approach be interesting for your new gpu system?

i am sure you work hard on this, and its just my two cents, but i dont see anything that makes your engine different from the rest - and i know your integration into max was always good - but for me it is not enough to change my pipline..

one last thing - and this is something i dont understand why other engines care so little about (except maxwell maybe - havent tried this) - the sky!
most use since 10+ years the preetham model - maybe with modifications, but still..
and they dont expose enough parameters (i read once there are about 50 variables?) anyway - mostly we get 3 and if we are lucky a horizon slider..
since years i fiddle with all the parameters and try everything possible to get some nice, not over saturated colors..
i know there are hdr etc - but having a flexible sky is imperative! ;)
..and most fail at this.. (terragen sky is very nice though)Would you say the same thing about VRay if it suddenly used the GPU not only for preview rendering (RT module....with lots of limitations), but the main rendering engine as well? It currently is CPU only. As is Mental Ray.

FinalRender R4 GPU isn't simply an equivalent to iRay or VRay RT. THE CORE RENDERING ENGINE IS GPU ACCELERATED. I don't understand how people keeping on missing this crucial point. Neither VRay or Mental Ray offers this, and you say you see no difference? Wow.

damjan
11-13-2012, 09:23 AM
it is the same in the sense that the main limitations of the other packages apply - even if the integration+shader support is better, gpu ram limit is a problem

Edwin Braun
11-13-2012, 05:25 PM
@Edwin - i am sure you know about the centileo render engine - what is it with the gpu chaching they do (accessing 10 times more ram than the gpu has, trough mysterious caching algorithms)
have you tried such thing? is it possible at all?

another thing - how about corona renderer - its still alpha - but it uses basically that what we know from vray - Lightcache, and then in the primary engine, path tracing with the lightcache data - and it is very fast (and cpu only)..
wouldn't such kind of mixed approach be interesting for your new gpu system?

i am sure you work hard on this, and its just my two cents, but i dont see anything that makes your engine different from the rest - and i know your integration into max was always good - but for me it is not enough to change my pipline..

one last thing - and this is something i dont understand why other engines care so little about (except maxwell maybe - havent tried this) - the sky!
most use since 10+ years the preetham model - maybe with modifications, but still..
and they dont expose enough parameters (i read once there are about 50 variables?) anyway - mostly we get 3 and if we are lucky a horizon slider..
since years i fiddle with all the parameters and try everything possible to get some nice, not over saturated colors..
i know there are hdr etc - but having a flexible sky is imperative! ;)
..and most fail at this.. (terragen sky is very nice though)

The paging approach is a possible method to use memory more efficient, however there comes a lot of restrictions with it. Our approach is to have a movie quality rendering system based on the complex nature of 3ds max and not a high performance special purpose rendering system.

CPU only rendering is also one approach one might choose, however to be effective the rendering engine has to be changed to massive parallel approaches (like GPU) We are working on it.

What do you want on sky ? We can see that users are even overwhelmed with ONE parameter/setting. One issue we have in fR is we do expose too much parameters.

-ed

AbnRanger
11-13-2012, 05:25 PM
it is the same in the sense that the main limitations of the other packages apply - even if the integration+shader support is better, gpu ram limit is a problemOf course, but you said there was no difference between R4 GPU and other renderers. There is a huge difference. Again, with VRay, you have a nice RT module, with all it's limits (no post FX, lack of certain shader support....SSS especially), but once you have your scene set up, now it's time to use the production renderer. It uses only the CPU. Same thing goes with Mental Ray.

What's more is that a solid case can be made that finalRender (R3.5) is already the fastest production render for Max. I've tested some scenes in both renderers (there is a nice Multi-Convertor available from the ScriptSpot), and finalRender is always significantly faster. Now you add to that GPU acceleration, and there's NO difference?

Edwin Braun
11-13-2012, 05:29 PM
Would you say the same thing about VRay if it suddenly used the GPU not only for preview rendering (RT module....with lots of limitations), but the main rendering engine as well? It currently is CPU only. As is Mental Ray.

FinalRender R4 GPU isn't simply an equivalent to iRay or VRay RT. THE CORE RENDERING ENGINE IS GPU ACCELERATED. I don't understand how people keeping on missing this crucial point. Neither VRay or Mental Ray offers this, and you say you see no difference? Wow.


Right, our trueHybrid™ mode is not really understood by many users. We did not release any information about it yet as this is one key feature we can see as a game changer for using GPU in movie productions.

I'll soon talk about this and show some videos explaining it a bit more.

edwin

damjan
11-13-2012, 05:53 PM
@AbnRanger - you are right if it is the case, that the gpu acts as a production renderer - and then it has to overcome all the limitations, memory, shader support etc. - vrayRT is a preview engine, which some people in product viz etc. can use as production renderer, but if you do something more complex, it fails.
dont get me wrong, i would love to see finalrender succeed here, but i dont see how it could without at least get rid of the memory limitation.
than again i never heard of the trueHybrid thing..

@edwin regarding the sky - i see your point.. its just that some people want an image with the click of a button..
and then there is people that do this 24/7 for a decade now - and they want a flexible, if possible vastly customizable program.. (which are for sure a smaller market)

anyway - its been a long time i tried finalrender, i remember it as the renderer that replaced lightscape for me (until others came along) but that's very long ago..
it was not my point to flame stuff that's not finished yet.. good luck to you finalrender guys!

CGTalk Moderation
11-13-2012, 05:53 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.