ATI's GDC06 presentation are up

Thanks for the heads-up. I always enjoy these documents. :)

Still waiting on nvidia...:mad:
 
Richard Huddy's presentation is so funny, he loves to talk about nvidia hw too :)
didn't know 7800 pixels batch size is about 100 pixels..
 
Jawed said:
http://www.ati.com/designpartners/media/screensavers/RadeonX1k.html

If you have X1k (or is it X1900 only?) then you can download and experience it for real.

---

The cloth demo (in the SDK) is pretty impressive.

Jawed


any ATI SM3.0 card can run that screensaver. Nvidia cards should work but dont, requiring a driver patch that they may or may not release. (BSOD on Nvidia cards). Lighting is certainly pretty but some of the textures on asteroids and planets and such are a bit low. Still crazy stressful for a screensaver.
 
Last edited by a moderator:
nAo said:
Richard Huddy's presentation is so funny, he loves to talk about nvidia hw too :)
didn't know 7800 pixels batch size is about 100 pixels..
According to Tamasi its > 200 quads!
 
Dave Baumann said:
According to Tamasi its > 200 quads!
And I think he's right. Since within a batch (but according their patents they call it a 'segment') every pixel shares the same progam counter you need more than 100 pixels aka passes aka clock cycles to hide texture latency.
 


Ahahaha, that is SO a replica of one of the scenes in the intro to Star Trek : Voyager!

I love you ATI... (now where's that heart emoticon when I need it).
 
Sobek said:


Ahahaha, that is SO a replica of one of the scenes in the intro to Star Trek : Voyager!

I love you ATI... (now where's that heart emoticon when I need it).
They also liked DS9 (intro again):



Conclusion: ATi devs really watched too much Star Trek.

(Not that this is bad though, bring us some more juice of that please. Looks awesome.)
 
Jawed said:
The cloth demo (in the SDK) is pretty impressive.

Thx! I wrote it. :)
This sample will run all the way back to Radeon 9500. The ps2.0 cards needs multipass for the simulation though, plus a final interleave pass (since it can only do one R2VB stream), but the cost of that isn't too great (it's pretty low-res so the most expensive part is the final rendering and not the simulation).
 
Hmm, interesting, the cloth demo looks anything but low-res!

I couldn't tell from the demo whether it detects self-collision - when the folding gets extreme it's hard to see, in amongst all the action, whether the cloth penetrates itself or folds around itself. Not sure what sort of complexity that would add.

I wonder if graphics programmers relish the chance to do "hardware" physics rather than "light bouncing around" physics...

Jawed
 
Back
Top