NVIDIA launches GPU based ray tracing at Siggraph
NVIDIA has announced that it can calculate ray traced imagery on the fly using its GPUs, claiming an industry first. Based purely on NVIDIA GPU technology, the ray tracer shows "linear scaling rendering of a highly complex, two-million polygon, anti-aliased automotive styling application." if you want to get down and dirty, then the image shown here was displayed at three bounces, performance was demonstrated at up to 30 frames per second (fps) at HD resolutions of 1920x1080 for an image-based lighting paint shader, ray traced shadows, and reflections and refractions running on four next-generation Quadro GPUs in an NVIDIA Quadro Plex 2100 D4 Visual Computing System (VCS).
Now, I know ray tracing is incredibly complex and calc heavy, but, really, if this is an industry first, shouldn't these images look better? I guess my point is that whether its full ray tracing, or fudged (some systems, like Autodesk Showcase, have some tricks to get over the calc hurdle), I think today's users expect more in terms of depth of realism.
Labels: GPU, Nvidia, ray tracing., showcase, visualisation
1 Comments:
Yep I agree Al, and this is a problem that we have encountered many times and over many years (!!) with a variety of tools and most recently encountered when we sat down with DS Vis team to put our heads together and debate rendering after we were acquired by them. Even with the best technology in the world, renderings and images can still look awful in the eyes of a designer if a few simple rules aren’t followed. I think the problems here are known, but worth mentioning once more as they always crop up.
a) The quality of data set. Here the quality of the data seems poor, yet the quality of the data used lends an awful lot of weight to the perceived final static (or real time) image. I would guess that the model used here is a poly model and not the proper Veyron surface model. A final release dataset adds a level of realism, from reflection lines to gapping. Our eyes (because we are used to high quality renderings and movies) pick up the fat highlights that dip where they shouldn’t, or the jagged flowing reflection around wheel arches and tell us that something is wrong…..
b) The ‘situation’ in which we see the car. A Veyron would be parked outside Mayfair, or an hotel in Monaco, or would be seen driving around Milan. When is the last time you saw a bright green Veyron in downtown LA with flat texture mapped buildings? Well never, apart from in games, which is cool. But in our world of design reviews and ‘in context’ visualization it somehow doesn’t work and our brain says ‘mmmm….something is not right here...’
c) The camera angle and set up. We, as designers, expect to see cool images, funky angles and subtle lighting. Composing and setting up a rendering is still an art form. For sure Bunkspeed and others (including the new Photoview 360 in SW) do offer short cuts and set ups, but you still need an eye for set up, camera angle etc. That’s still a skill that that can take a while to learn.
So all told, if we get the above wrong it means that even when we look at an image with all that cool technology under the hood we say can still say ‘err right, but it looks awful and not very realistic!’ I think this is the issue here, get the rendering fundamentals rights and its happy days. Even the world’s simplest rendering tools can be made to sing as long as basic rules are followed.
Post a Comment
<< Home