Worley Labs will release a new product called “FPrime” soon. The rendering speed is almost real time!
http://www.worley.com/cool/FPrime_texturing.avi
http://www.worley.com/cool/FPrime_speedTracer.avi
Does anyone know what kind of technique is possible to increase such “high-speed” rendering ability?
There is certainly a lot of work being done (esp. in academia) on offloading the raytracing process onto GPUs (i.e. the specialised microprocessor on board graphics cards). There are a significant number of reports about this on the ACM Digital Library site (portal.acm.org) - though I think you need to be a member to see this…
Also, Purcell et al have produced a very fast Photon mapping solution using GPU’s - so this expensive technique (and others like it) could soon become substantially more commonplace.
What I am wondering myself, is whether the likes of Pixars Renderman will start to take advantage of this GPU-offload procedure - and if so, would they use it to decrease render times, or to produce RADICALLY more realistic and amazing results in the same time as before? Interesting stuff!
It’s a render preview rather than a renderer. Like Maya’s IPR or Pixar’s Irma (except loads better, by the looks of it). It’s not really real-time, and it’s definitely not real-time raytracing in the strictest sense of the word.
The work on photon mapping on the GPU is certainly interesting, and I’m certain that there is software in the works that will take advantage of this… but we won’t see anything really useful (for high-end applications) being done on the GPU until the next couple of iterations at least. Apparently 1GB cards are not far off. When we get a better programming language than Cg in its current state, and these new cards, along with renderers to utilize them, then we’ll be laughing.
This plugin seems not using GPU to render the preview, the requirement for display card is any video card. Some Maya users told me that IPR has a lot of limitations and slow, compared with FPrime.
from their webpage:
" FPrime does not use graphics hardware to render, so it has no light type or count limit."
“A basic speed and quality comparison with LightWave’s renderer, starting with LightWave’s own Raytrace benchmark scene. Notice FPrime renders a final antialiased image in a fraction of the time of LightWave’s own renderer without any AA at all!”
I don’t see anything saying it’s a preview renderer.
I agree. It may have the benefits of IPR etc (may be even better) - including ‘real-time’ feedback when setting up lighting schemes, but it has the added benefit of producing full-quality renders.
One thing I notice from the videos, however, is that the rate of improvement of image quality seems quick at first, adn then slows doen as the image gets progressively more detailed… I wonder how long it takes to produce something which matches the quality of Lightwave render, on average (rather than for a select number of test cases)…?
I think maybe they mean, by not needing specialist graphics hardware, that they don’t need ray-tracing accelerators such as ART - they talk about the graphics hardware not limiting the number of lights etc. Using the GPu to aid raytracing certainly would not place a limit on the number of lights in the scene…
… I could be wrong through! :shrug:
One thing I noticed is that it only updated the areas of the image that change. If you watch the vid with the car where he’s changing the lighting, you can see that only one side of the car is changing.
Anyway, real time ray tracing is totally possible. In fact, if you don’t do any sub-pixel sampling where you’re shooting more then one ray, it’s actually quite easy. Most of the work would go into optimizing the scene though. Multi sampling is usually what really slows you down.
Here’s my current fav in this area: http://www.openrt.de/Gallery/
These guys are seriously insane! look at the real-time global Illumination.
-=GB=-
You guys might want to take a look here:
http://www.realstorm.de
These are demoscene guys who made a realtime raytracer. I recommend the “Realstorm benchmark 2004” and the demo “Still Sucking Nature” as downloads. Not just technically but also artistically impressive. 
I see that they are using radiosity, which is not at all like what most renders use today, most use monte carlo or photon mapping. The great fun of radiosity is that the estimate that is used can easily be computed by a GPU, as it is essentially projecting polygons on surfaces. I do not know what speed this would result in, but I can imagine it will be impressive.
Originally posted by playmesumch00ns
… When we get a better programming language than Cg in its current state…
We already have GlSlang, which is a more accesible way for professionals. This with Ueber_Buffers should do all the work we need.
cu
Tom
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.