Saarcor at SIGGRAPH 2005
- Thread starter Xenus
- Start date
damn i hate how siggraph always sneaks up on me. This RPU sounds interesting, any links?
epic
epic
Megadrive1988
Veteran
maybe ATI will acquire Saarcor and develop its technology to the fullest extent, to become the R900 or R1000, and be the architectural basis for Xbox3's graphics.
I can dream can't I
I can dream can't I
RPU? Oh, noes...
I expect at least 2 topic a day on the Console forum relating rumors that the Revolution could be embarking a RPU.
Anyway, a Ray Tracing unit would be useful only if it's accelerate GI algorithms like Photon mapping or Radiosity.
A plain vanilla Ray-traced image would look a lot worse than an image from a classical renderer (using fragment operations).
Sure, the reflections will be exact, without any warping, refractions too will be great, but that would be about the two sole positive points.
I expect at least 2 topic a day on the Console forum relating rumors that the Revolution could be embarking a RPU.
Anyway, a Ray Tracing unit would be useful only if it's accelerate GI algorithms like Photon mapping or Radiosity.
A plain vanilla Ray-traced image would look a lot worse than an image from a classical renderer (using fragment operations).
Sure, the reflections will be exact, without any warping, refractions too will be great, but that would be about the two sole positive points.
davepermen
Regular
a raytracer in the basic (witted) setup would look about the way doom3 looks. (even if you enable softshadows in doom3).
this is because doom3 got to about the same featureset. yes, the refraction-effects on the windows would be correct instead of fakes, and reflections, too.. but the main lighting effects would be of same quality:
direct illuminated pointlights with sharp shadows.
then again, a hw that can trace rays (and has shading capabilities) is 100% programable for any form of gi, be it path-tracing, montecarlo style, or photonmapping. possibly even metropolis lighting transport
the new hw version 2 they'll show looks really nice. one such card, and one physics card in my cpu, and i'll be happy
this is because doom3 got to about the same featureset. yes, the refraction-effects on the windows would be correct instead of fakes, and reflections, too.. but the main lighting effects would be of same quality:
direct illuminated pointlights with sharp shadows.
then again, a hw that can trace rays (and has shading capabilities) is 100% programable for any form of gi, be it path-tracing, montecarlo style, or photonmapping. possibly even metropolis lighting transport
the new hw version 2 they'll show looks really nice. one such card, and one physics card in my cpu, and i'll be happy
davepermen
Regular
Vysez said:
one main point that i'd see is a much more stable framerate than doom3 shows on a rastericer. in doom3, even with tons of tweaks, those shadows can sometimes kill fps. in a raytacer, the fps would be rather constant. not the highest fps, but not the lowest one as well.. this is a cool thing.
imho this is the main positive point actually. it shows that in raytracing, you don't rely on complex hard to predict hacks to get acceptable results. thus, no terrible fps breaking if you f.e. walk over a floor wich gets some shadows from a grid at the ceiling.. this would yield tons of fullscreen quads drawn on a rastericer, but not any more shadowrays on a raytracer... result: rastericer goes down terribly. raytracer just goes on..
Can anyone speculate why the second prototype is running at a lower clock speed than the first and, I assume because of it, running similar benches like the UT level slower?
EDIT: well, a reason for the slower performance seems to be offered in the PDF (programmable vs fixed architecture).
EDIT: well, a reason for the slower performance seems to be offered in the PDF (programmable vs fixed architecture).
Megadrive1988
Veteran
ok, within the next 7 years, in time for Xbox3 and PS4.....
what is "the way forward" as far as reaching a new goal for realtime graphics for consoles that will cost $300 to $400? it seems that classic Ray Tracing is NOT the right approach, for many reasons. so what about Global Illumination? Photon Mapping? I am expecting some sort of hybrid renderer to be used. meaning that, it is still somewhat based on scan conversion.
ok i am clearly out of my league as far as 3D tech-knowledge goes, I feel like I do not even know WTF I am talking about, or trying to say.
except that, I want to know, what is likely to be "the way forward", not for the distant future (10 years out or more) but for the mid-term, that is, in this case, well *less* than 10 years away, well within the time that Xbox3 and PS4 will be out. around 2011-2012.
what is "the way forward" as far as reaching a new goal for realtime graphics for consoles that will cost $300 to $400? it seems that classic Ray Tracing is NOT the right approach, for many reasons. so what about Global Illumination? Photon Mapping? I am expecting some sort of hybrid renderer to be used. meaning that, it is still somewhat based on scan conversion.
ok i am clearly out of my league as far as 3D tech-knowledge goes, I feel like I do not even know WTF I am talking about, or trying to say.
except that, I want to know, what is likely to be "the way forward", not for the distant future (10 years out or more) but for the mid-term, that is, in this case, well *less* than 10 years away, well within the time that Xbox3 and PS4 will be out. around 2011-2012.
I think that classical renderers will be with us for quite a while still, unless some unexpected revolution happens. Eventually, they`ll be so similar to a general purpose architecture that you could theoretically implement all of the things you quoted on them, but I don`t think a shift towards a different way of handling graphics-eg. a dedicated raytracer-will happen. Will GPUs be programmable and powerful enough in the next 7 years?I can`t tell.
