On 7/12/06, Petter Urkedal <[EMAIL PROTECTED]> wrote:
On 2006-07-12, Timothy Miller wrote:
> Should I infer from the silence that everyone thinks that the last
> draft I posted will get accepted? :)
I think it looks good. I will just second a few points...
o Why closed-source drivers are a problem
In my opinion stability is the most important point. When using a
proprietary driver for my current card, I have about 2 crashes a month.
That's weird to me, since I know the distributors pick out well tested
and rock stable kernels. Someone who is new to Linux may not have the
same expectation, and may never try to disable the proprietary graphics
driver, so Linux gets the blame.
We need to remember to include this in the full speech.
o No programmable shaders, T&L done in software
Just curios, since vertex shading is in software, it can be
programmable, right? So, a fish-eye or panorama lens is doable?
Absolutely! (Although I don't know exactly what Mesa supports.)
BTW, I think this is also an important point, recalling the "we just
want 2D" discussion. OGA is smart in a modest way. And I think it's even
future-proof. It may take 10 years before CPUs can do what graphics cards
do today, but as the cores cell-divide, the OGA approach becomes more
feasible even for heavy 3D. Other GPUs may move towards more general
purpose vector processing. Maybe we'd first make an independent
co-processor board for scientific computing and, of course, for further
accelerating the OGC1 driver. The same core could eventually end up in a
future OGC card. Okay, I'm diverging from the talk, people will probably
want to hear about things that sound realistic ;-)
Well, OGA, as a concept, has lots of future-proofing, as long as we
stay away from high-end games that require programmable fragment
shaders. As CPUs get faster, the vertex shading will require less CPU
time, and as busses get faster, the time to transport commands via DMA
will shrink.
Some day, we'll do programmable shaders, but that may very well be a
fork in thought, rather than an entirely new direction.
Fixed-function rasterizers will continue to be all you need for
desktops for a long time to come. I'm guessing that Aero Glass and
DX10 require it as a way to offload additional work from the CPU to
the GPU, but I haven't seen anything in screenshots that couldn't be
done very nicely without programmable shaders.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
