still wondering which future for rendering?

http://www.cgarchitect.com/news/SIGGRAPH-2009-CHAOS-GROUP-GPU.shtml

vray RT implemented on cuda.

the gain in rendering time (with same quality) of a SLI nvidia compared to the 8 core cpu is simply amazing, going from one minute to a couple of seconds.

Cuda is nvidia-based, but OpenCL not: the future of rendering is there, for sure. :slight_smile:

i hope ton, brecht and the others at siggraph attended at this presentation…

Impressive!
Future is there for sure. Maybe after photon mapping & FG, pathtracing and BDRF shading, the next big deal for internal renderer will be moving to OpenCL…

I always say, why wait…when something is obviously this much better make the move. I f there is no support create the support…with that being said…I am not experienced with blender code, let alone digging into the nuggets of a renderer…

Soooo, I have to go and by another card to take advantage of this? Until its standard kit count me out.

There’s also this.

I’m actually considering buying V-Ray(~$1000.00) just so I can use this. I imagine end users can be a lot more productive without having to wait for long render times.

let’s hope OpenCL will be as good as Cuda (a)

If OpenCL is going to be as great as CUDA (which I hope after some development time) then we should donate for OpenCL integration in Blender!!! I would definitely spend some money on such an awesome technology!

I assume in case man power is there that the GPU move when openCL is done will come as well.

but having openCL integrated to speed-up GI-less renders sounds odd to me…

:smiley: Nice zinger!

Won’t OpenCL also allow acceleration of other things as well? Could fluids be calculated on a GPU with OpenCL?

You’re saying your blender machine doesn’t have any kind of gpu based graphics card in it whatsoever?

If you don’t find this impressive, you’ve missed several key points:

  • They’re offering orders-of-magnitude faster rendering. Not just 10%, 20%, but 8 to 20 times faster.
  • They’re doing it using commodity gaming cards rather than expensive custom gear like the defunct art-vps systems.
  • Unlike gelato its matched to an existing CPU equivalent, so studios that have renderfarms that aren’t GPU capable (ie most blade and 1U rack systems), they’re not locked out.
  • They’re planning to port to openCL so it won’t be tied to nvidia or ati; any gpu will be used, and with gpu performance year-to-year currently outpacing cpu’s, end users will see substantially faster results for less money in less time.

So ideally your lighters will have a $100 gpu graphics card, and they can preview full multibounce GI in near realtime. Get through the lookdev side quickly, then send it to your existing renderfarm for sequences.

As even the vray developers hint at though, its not really worth any blender developers time to even touch this until openCL is finished and ready to use. But its an amazing sign of things to come.

I’m currently working on a commercial involving giant photoreal robots; I thought I was doing well to get away with 2 minutes a frame with a physical sun/sky light and final gather. I then watched that demo, and my heart sank. C’mon openCL, I want the impossible NOW!

-matt

Quoted for agreement. I hope somebody… will read this and will do some magic.

Heh, thanks, but just to make absolutely clear:

its not really worth any blender developers time to even touch this until openCL is finished and ready to use.

-matt

real pumped for this! Vray already is super fast compared to any other renderer at the same quality. the near realtime production of those renderings seems like something out of a dream… would be REALLY cool! no more overnight renderings and then having to re-render because of some tweaks.

Anybody reading this as well?
NVidias own interactive rendering platform.
CUDA again (of course).

http://www.nvidia.com/object/io_1249366628071.html

http://www.xbitlabs.com/news/video/display/20090804145325_Nvidia_Develops_Interactive_Ray_Tracing_Engine_for_Graphics_Chips.html

Cg, Stream and CUDA are proprietary (no go). BrookGPU is hibernating and not as efficient on OpenGL. Lib Sh is no longer maintained. OpenCL isn’t there yet.

In short, no good, open GPGPU platforms are available.

OpenCL is not mature yet, that’s for sure.
But it doesn’t mean we have to wait before starting thinking about GPGPU rendering.

Nvidia when acquired Mental Images had a great plan for shifting the rendering to the GPU and they are already doing it.

And Mental Ray, with awesome integration with RT shaders (Cg), Linux, MacOSX and Window$ versions, is really running ahead the competitors (even Vray: i bet in a couple of years the MetaSL materials will be used everywhere).

if you see the VRay RT demo you will believe that openCL and similar approaches are where things will go to.

I also agree that openCL is in infantsy while maturing fast.