Wow i’m impressed. Regardless the number of gpus involved, there’s A WAY to get that result in almost real time that wasn’t available before. Now, evolve RT to that and I’ll be happy 
I see a future where RT window will be inside NUKE or FUSION and u’ll be “comp-rendering” without thinking in “oldschool-rendering”.
This is not directly reported to iray, but it’s my old test for VRay-RT in a heavy scene.
6500 teapot (6.5 Million poly) (not instance, but simply copy geometry), all Glossy, 50 IES light with Areashadows + GI + DOF.
Plus a heavy geometry deformed in “realtime” with a modifier.
The slow part is to copy the 6500 teapot from 6 Million to 12 million ( 2 minute). But after that, VRay-RT works very well. .max file is 60 MB with 12M poly. Main PC is a dualXeon 5440 @3.250 Ghz 16GB + 2 node i7 @ 3.8 Ghz 12 GB.
I’m waiting the GPU version.
You can get the same performance from V-Ray RT right now through distributed rendering.
Best regards,
Vlado
What I basically meant was that I’m waiting for GPU-accelerated Vray RT. I’ve been testing our RTs with plenty of computers in DR (7 or 8 dual quad cores) and feedback is not quite as good…especially in interior rendering (outdoor rendering, especially with VRAY Domelights, flies at full speed). Obviously we are talking about cpus and not gpus.
On the other hand, can anyone tell me what’s that application they are using on the internet browser to control different rendering parameters, like time of the day and such? I can barely hear what the guy is saying. It looks like the ultimate client feedback application, though.
I thought one of the main problems was is it really cost effective to outfit a bunch of render slaves with GPU’s? isn’t that one of the reasons gelato failed.
It ain’t, because you have to outfit said renderbox with a good CPU and a high-end video card, which is not cheap at all.
VRayRT GPU (as shown at Siggraphy) however, isn’t pulling a “you need a quadro buddy”.
I was thinking that instead of looking at a computer in the 3000-5000 euro range of prices, we should consider creating a computer+processing cluster in the 15.000 - 20.000 euro. That would allow client-assisted rendering sessions with acceptable lag. I was thinking about this as an analogy of Autodesk Flame and inferno, both solutions available in hardware-powered flavours, which cost (or costed) about 300.000 euros and 600.000 euro, respectively.
It sounds like a LOT of money, but specialized facilites were happy to pay it and crunch over 2K sequences effortlessly, mainly in the advertising postproduction field. Now if we think about the future of 3D interactive render, maybe we should start considering clustered DR nodes as a default configuration for any given “hero” workstation.
I’m not a hardware guy…and i already mentioned this…but does this mean that the machines that have multiple video cards…like mine which currently has two Nvidia cards in it with room for a third…would make for better machines to have as far as the real time stuff is concerned?
Same thing, but different 
VERY different. Requires it’s own shaders and workflow.
Hell.
No.
Just to be clear, Iray is NOT a realtime version of Mental Ray!
It is a GPU based rendering system in and of itself, what you
see is what you save out. So you wouldn’t use Iray to do set
up work and then render out with Mental Ray as the results
will be different. This is a horse of a different colour as it were.
Regards
Bri
this is very impressive. it looks like things are going to get alot faster in the not so distant future. the next generation cuda architecture (Fermi) was recently announced http://www.nvidia.com/object/fermi\_architecture.html
The new iteration has 512 CUDA cores copared to the current 240 cores on the tesla. Fermi is designed for C++ which will make it available to a wider range of apps.
though these interactive demos look cool, i still am not seeing what i think alot of poeple are looking for and that is the ability to use this technology to render out a sequence of frames. surely that is in the works on both the v-ray and mental ray sides…its the most practical application for this imo.
this is an important issue that anyone looking to get a renderfarm needs to consider. it looks like you need to make sure your servers have the ability to incorporate a GPU in the future.
i asked Ken Pimentel at Autodesk if we should be recommending GPU ready servers to our customers…his reply…
“The things we’ve been investing in 3ds Max (the most GPU-centric of any of the professional 3d apps) will pay off big time with this new architecture. (referring to Fermi) You should be putting in the most powerful GPU customers can afford (lots of video RAM!) because more surprises are coming from us. Your customers will thank you.”
so it looks like its coming, just not sure when. i have a feeling the next release of max. and this isn’t just related to rendering. non rendering related tasks can benefit from the GPU as well.
-joe
Traditionally people that wanted the most power bought dual processor workstations, now it looks like GPU’s will be taking over as the primary rendering hardware so does that mean we no longer need these dual processor systems for rendering?
be aware that this opinion is rather biased, but then again he knows the inside and outside of mental images…
http://forums.cgsociety.org/showthread.php?t=812125
http://forums.cgsociety.org/showpost.php?p=6138458&postcount=54
Well well,
I can’t really let pass a new mental announcement without some constructive deconstruction… so:~ Ahhh… yet another half-baked mental “solution”! ~
Proof that it’s half baked: none
History that backs me up: MetaSL, mental mill, reality server, mental queue, mental etc. (mmm maybe someone should patent this last idea ).But let’s move to some further thoughts:
* iray in reality server
- At NV’s gpu conference iray is shown embedded within reality server. Why?
Simply because mental is still trying to find a way to market reality server, which is their biggest product failure as it is basically unsold as of today so they are putting inside anything they have to make it appealable. Truth is that one can simply build rs just by using opnesource softwares (see old threads). For who is still confused about rs (it’s not your fault, just mental’s) it is basically a pipeline to do sever side rendering with remote users being able to control and tweak certain parameters. All packed in a ultra awkward, complicated - hey let’s use the right word for one time! - “mental” pipeline. It includes several software and hardware renderers, iray being one of the hw ones.- iray in sceniX: We will see if how it gets in.
Also mental mill is integrated in FXComposer, and it wasn’t a revolution for the industry…* iray
First of all: sincerely congrats to mental for having restrained from putting in the faq that iray is for production usage. Then:
- developed by mental, NOT by nvidia (hence the sadness)
- does not use optix api (optix if you don’t know is not a renderer but an API, it is the official name of the old codename “nvirt”: nvidia raytracing api). Hint: further proof of the how few a close-minded company as mental collaborates with an open one as nvidia (this is because of mental management, not nvidia). The orange bugatti shown at the start of gpu conference talk is optix:
http://www.youtube.com/watch?v=vPeZ...feature=related
and this is also optix:
http://www.youtube.com/watch?v=BAZQ...feature=channel
- closed: Optix API will allow developers to write their own “iray”, in a more open and controllable fashion (when Optix API will be released you will see for yourself the API and code examples) again
- hardcoded shader (based on archmat) and no programmable shaders (yes, not even MetaSL! hint: even mental does not use its own technology…)
- no motion blur
- no curves (so no hair nor fur)
- no particles
- no volumes
- no SDS
- no NURBS
- no displacement
- no sss
- performance: see for yourself results on 15 GPUs
hint: mental removed images from PDF…- hint: the scene used for demo features a very large horizontal window, on a more tricky setup, with smaller light entrance noise should appear and converge-time increase
- no bidirectional correlation to mental ray (so you basically need a double pipeline)
- not useful for relighting: if you think you can use iray to relight your mental ray setup and then switch back to mr for final rendering, well, this is not the case.
- misleading faq: mental ray 3.8 integration: iray is not “integrated” with mray 3.8, it will be “shipped along”.
- integration on autodesk softwares: I would be very surprised if they do it, it has no value for 3dsmax/maya/softimage. Hint: no words from 3dsmax prod managers in this thread.
- No non-nvidia gpu, no OpenCL support
- mental is again late. VRay RT has at least 1.5 years of advantage.
* MetaSL
- still no VM compiler! Hello mental! 5 years have passed. And now it’s too late: RSL2.0 and OSL.
- metasl shaders support in mental ray: slow, unstable, not full feature
- showstopper performance issues for compilation of real production shaders
- no adoption in film industry
- oems: afaics only autodesk and dassault are using metasl and only for HW previews (as I said there are serious performance issues for on-the-fly compilation of real shaders)
- metasl has probably a very limited future because of RSL2.0 and the upcoming OSL (which will provide all infrastructure, including compiler, opensource).
These are just my opinions anyway, I guess now is time to just sit back and enjoy the usual nice comments
edit: I also find ridiculous that specially with mental products the symbol of TM or (c) or (r) is omnipresent, even on thread title… It definitely shows up some mental attitudeTM.