Nvidia CUDA - the future of rendering?

[blumentopferde]

Hi I've been stumbling over a few new renderers using Nvidia Cuda, that claim to be so fast that you can actually work in the render preview window. Here are the links:

 

http://www.refractivesoftware.com/

http://www.randomcontrol.com/arion

 

Whenever there's a new release of a Renderer you will read press releases praising the speed of the renderer but when you try it yourself you will find yourself growing old waiting for results, so I am sceptical...

Is there anynone who already tried a CUDA-based renderer and can confirm if it lives up to its promise? Does Anybody know if these Renderers are compatible to material libraries for other 3D-Software and Renderers? Did anyone had the chance to compare speed and performance with common renderers?

Would be great to hear from your experiences!

[braddewald]

I'd be interested to hear about other's experiences as well.

[AJLynn]

Waaaaay too early for statements like "the future of rendering". And I think that as this sort of thing matures you're going to be seeing it moving away from proprietary technology like CUDA and into open standards like OpenCL.

[adarsh]

Until a standardization is agreed upon, there will be many independent verticals of application / rendering support with respect to the hardware.

Cell Processor -- IBM/Sony/Toshiba

CUDA -- nvidia

Stream - ATI

OpenCL, OpenGL, Directx (Autodesk QuickSilver - cpu/gpu)

 

The problem being not all aspects of the scene can be rendered using GPU, a part of the scene has to be handled or rendered by the CPU.

GPGPU (Larrabee from Intel: where in application can be used as is without porting to GPU, this is still in research). The major part is to

recognize which part of the scene can be handled by the GPU (embarrassingly parallel) and CPU (sequential / parallel).

 

Would be interested and enthusiastic to learn more about this.

 

Thank you,

Adarsh

[divanovic]

I would agree on being to soon to call it a future of rendering, because of serious lack of possibilities still.

You may try 'furry ball' rendering engine, which has a 14 day trial, but you'll notice rendering quality being far from standard, imho...

Apart from that, hardware has never been cheaper and render times are not such a big issue anymore. For less than 50k$ you may get equal to 40 xeons from Boxx, for example, and that figure is really nothing for large companies today. Both for film and game making in which I think Cuda will have more chance to prove.

 

just my 2cents,

cheers

Dejan

[blumentopferde]

Thanks for your opinions!

Well, if you're a student or a one-person-company 50K$ is still one extremely large figure... I'm thinking (or dreaming) of having renderings done with a standard PC or a laptop in short time. Being able to quickly compare several variations of lighting, textures and camera settings without having to invest hours or even days. Or to be able to run through a 3D-model like a photographer and take snapshots of interesting perspectives. If the speed increase is really as big as promised one might be quite close to that...

[AJLynn]

Ask and I shall receive. The upcoming Vray RT for GPU is OpenCL. Alexander, that software could be used for your "snapshots" - but the output quality is not as good as regular Vray, so you're still going to want to do your real rendering on the CPU.

[branskyj]

Well, Octane renderer (Refractive software) is in fact a production quality renderer, completely GPU based and it does deliver what it promises. I bought it myself (it is dirt cheap), it has remarkable technical support and is being updated every month. The only problem I have with it is the complete different work flow that I have to follow. I am used with the 3DS Max + VRay method and adjusting to the new work flow was not easy.

 

From what I saw the new VRayRT(GPU based)will be much much better than the VRAyRT (CPU based) in terms of render quality and speed. And good news- it is OpenCL which means not only NVidia but ATI as well.

[StompinTom]

OpenCL is actively being developed as something that will make GPU computing available to non-Nvidia cards.

 

SmallLuxGPU is a proof-of-concept OpenCL GPU accelerated version of Luxrender (luxrender.net) which is an open-source unbiased renderer, much like Maxwell, Indigo, Fryrender and now Octane. They've gotten some very impressive speeds both with Nvidia and ATI cards.

 

The thing is, CUDA is far more mature than OpenCL right now, so it's taking the lead as far as desktop supercomputing goes. All I can say is that the next few years will be interesting.

[AJLynn]

I don't know about taking the lead. Adobe is currently on CUDA for the Mercury feature, as are these two things here, but Quicksilver, VRay RT and Apple Aperture are all cross-platform. nVidia did get to market first but by introducing proprietary tech they didn't do themselves any favors in the long run. I think it looks pretty clear that OpenCL is going to take over the market very quickly.

 

This is a good thing. nVidia and ATI need to keep switching places as the hardware leader in order to keep pushing the technology forward so the GPU computing platform can't favor either.

[Guest zarfworld]

Sorry to be so naive but what is OpenGL?

[AJLynn]

Not OpenGL. OpenGL is this: http://en.wikipedia.org/wiki/Opengl

I was talking about OpenCL, which is this: http://en.wikipedia.org/wiki/Opencl

[blumentopferde]

Indigorender is jumping onto OpenCL and CUDA, bunkspeed is jumping onto CUDA.... hope that Maxwell, my favourite, will follow this trend...

 

http://www.indigorenderer.com/content/announcing-gpu-acceleration-indigo-renderer

http://www.bunkspeed.com/shot/

 

As far as I can read from the announcements CUDA is absolutely taking the lead... OpenCL will definetely not take over in this or the next generation of GPUs - one has to consider that most of the announced Software won't be finished before the next generation of graphic cards...