GPU accelerated rendering, why only that?

loocas | miscellaneous,opinions | Wednesday, January 13th, 2010

It seems that GPU accelerated rendering has been the hottest topic recently. But why should only the rendering get accelerated on the powerful GPUs?

There are tons of other applications that desperately need acceleration. Simulation for example. Cloth, hair, particles, rigid bodies etc… all need some heavy calculations and are actually quite simple, in comparison to rendering that is. Wouldn’t you prefer your cloth sims to be faster than real-time? Wouldn’t you love to be able to have physically accurate dynamics in your particle simulations? Or better yet, wouldn’t you love to be able to mix all this together in one mighty-powerful framework, all running on our GPUs? I would!

I understand rendering is the hottest shit. It provides results both for the clients as well as for the developers and their PR. It’s cool, sexy, in, fresh, new, first and what not, but I’d really much rather see some other GPU utilisation than just viewport acceleration and some rendering attempts or, even worse, real-time rendering previews, such as the VRay RT etc…

The thing is, when you render, it’s still a semi-product, it’s not what the client will/should ultimately see. The rendering has to go through at least a basic post-process (color correction, perhaps simple A over B composition or similar). The thing is, the client most likely won’t be sitting at your workstation watching over your shoulder whatever you’re rendering. No! But they will art-direct your sims for example! They’ll tell you “ok, add more in here, but make it blow from upper-right and give it a bit of a punch on this frame”. They will want you to make the clothing more flappy, less flappy, more rigid, less sticky etc… But rendering? Nope. You’ll send your batch on the farm and have it rendered, stored on a framestore (or wherever) and have it delivered to the compositor, who’ll then take those frames and add unicorn tears and fairy dust to the comp so that the director/supervisor can see what the hell you’ve been spending your weekends on.

That being said, I’d too love to see some GPU accelerated rendering, at least partially (if it is even possible), but quite honestly, much more useful and helpful would be to see other areas utilizing the amazing power of modern GPUs, especially when Fermi is coming soon.

4 Comments »

  1. Nvidia gives away a PhysX DCC Plug-Ins for Maya and 3DSMAX

    http://developer.nvidia.com/object/physx.html

    Comment by steve — January 14, 2010 @ 03:23

  2. Hey, steve, yeah, that’s right, but as far as I know, Physx only accelerates Havoc in 3ds Max (don’t know about Maya though), which is a set of dynamic solvers, including cloth, rope etc… but what I was talking about was rather to be able to offload the calculations of ANYTING ELSE on the GPU. For example, regarding sims, the Cloth modifier is a great cloth simulator, but slow and single threaded. If that ran on a GPU, that’d be sweet!

    Comment by loocas — January 14, 2010 @ 10:39

  3. It is difficult to map dynamic solvers efficiently to a graphics card, after all there isn’t much in the way of multi-threaded dynamics (beyond particles), you need to be massively parallel to get full effect of a graphics card. You will likely particles and fluids go GPU accelerated first in any of the major packages.

    But first we will have to get past this GPU rendering/compositing phase.

    Comment by John — January 14, 2010 @ 11:54

  4. couse you focus on rendering – http://www.nvidia.com/object/cuda_home.html#state=home

    Comment by martincg — January 24, 2010 @ 19:37

RSS feed for comments on this post. TrackBack URI

Leave a comment

Powered by WordPress | Theme by Roy Tanck