Moin,
Send only to me, thus full quote.
On Tue, 17 Apr 2007 10:24:00 +1000
Hugh Fisher <[EMAIL PROTECTED]> wrote:
> Attila Kinali wrote:
>
> > Using a on board general purpose CPU on the graphics card will
> > not give you any advantage at all. If a PC CPU is too slow, how
> > do you want to beat that with a CPU that you can put onto a graphics
> > card without implementing half a PC on it?
> > No, the only reasonable way to do that is to optimize the hardware
> > for this specific task.
>
> While I agree with Attila for the specific case of video
> decoding, the second generation OGD graphics card will
> need a CPU to implement programmable shaders, so it's
> worth starting to think about it now.
Although you can see shaders as a form of CPU,
they are a very specialized form of CPU. And they
have to be specialized to achieve the performance needed.
Thus i would not talk of shaders as CPUs.
> A GPU isn't a true general purpose CPU, and it will always
> need some specialised hardware support for texture access
> and pixel operations. To implement modern (2.0+ or 3.0+)
> shaders it does need floating point registers, a float ALU,
> compare and branch instructions, some bitwise integer ops,
> float to/from int conversions, ... so while not a true
> general purpose CPU, it's not far off either.
>
> nVidia and ATI have designed specialised CPU ('shader units')
> for their cards. I think it is reasonable for the OGF to
> consider using a general purpose CPU on the card because it
> will be quicker and easier than designing our own. We'll end
> up with a bigger chip than we need, but we know it will work
> right away and come with an assembler and debugger.
No it will not. Shaders are highly parallel SIMD units.
general purpose CPUs are SISD constructs with SIMD extensions.
Very few of them can execute more than 2 SIMD instructions
in parallel which limits the number of shader operations
done in parallel which in turn limits the fill rate
(super linearly because there are second order effects
that are not obvious).
Not to talk about the limited I/O bandwidth and added
latency that you get when going off chip. And even less
about the issues when you have more than one system
accessing the memory (which you can get around if you
have full control of them if they are all on the same chip).
> I didn't think ARM chips were very good at floating point,
> but maybe that's changed. My understanding is that you can
> buy ARMs as a logic cell to drop into your chip design -
> but I'm a software guy, so couldn't say how much of an
> advantage that really is.
>
> Other choices would be a PowerPC with AltiVec, or a MIPS.
>
> --
> Hugh Fisher
> DCS, ANU
>
Attila Kinali
--
Linux ist... wenn man einfache Dinge auch mit einer kryptischen
post-fix Sprache loesen kann
-- Daniel Hottinger
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
