Unified Shaders

[mapel110]

“It’s not clear to me that an architecture for a good, efficient, and fast vertex shader is the same as the architecture for a good and fast pixel shader. A pixel shader would need far, far more texture math performance and read bandwidth than an optimized vertex shader. So, if you used that pixel shader to do vertex shading, most of the hardware would be idle, most of the time. Which is better – a lean and mean optimized vertex shader and a lean and mean optimized pixel shader or two less-efficient hybrid shaders? There is an old saying: ‘Jack of all trades, master of none’,” Mr. Kirk said in an interview with ExtremeTech web-site.

Also wird das ganze ineffizient?
Quelle (http://www.xbitlabs.com/news/video/display/20041223075525.html)

[Quasar]

Vertex-Shader sind ggü. ff-HW-TnL auch schon ineffizient.

Es ist AFAIK eigentlich immer so, daß mit zunehmender Spezialisierung auch die Effizienz steigt.

[Demirug]

Kirk war ja schon immer ein Meister wenn es darum ging mit seinen Aussagen Verwirrung zu schaffen. IMHO muss man das ganze sehr vorsichtig lesen und sollte nicht versuchen daraus zu viel abzuleiten. Krik interpretiert die Wahrheit auch noch wenn es den entsprechenden Chip schon lange gibt.

Aber um seinen Gedanken weiter zu spannen. Arbeiten zwei Jacks schneller als ein Master?

[Coda]

Vertex-Shader sind ggü. ff-HW-TnL auch schon ineffizient.
Bei R300 bin ich mir aber sicher (nach ATi Doku), dass es gar kein festverdrahtetes T&L mehr gibt.

[mapel110]

http://www.beyond3d.com/forum/viewtopic.php?topic=19258&forum=9
We asked ATI’s Eric Demers, from ATI East Coast (Santa Clara) Engineering / Design center what his thoughts on David’s comments were:

“I pretty much agree with David Kirk's answer for this question. There's no requirement to build a unified shader architecture for the future, yet. It's more of a suggestion, in terms of implementations. Of course, all the shaders do share a core functionality that is identical. But the different shaders require different features on top of that (which are pretty different), such as texture processing.

But if one were able to figure out a way to unify the shaders such that nothing extra is required to fulfill all the requirements of the different shaders, while being able to share all their commonality, that would be a great solution. It would allow for optimal performance with a fixed set of resources, with any kind of shader load. With a unified core, you could always use all your resources for one thing; in that case, a heavy geometry app would run much better than on other HW solutions (no, there aren't many heavy geometry apps now, but who knows in the future). As well, scalability of overall performance would be better.”


hört sich schon etwas anders an.