Timothy Normand Miller wrote:
On 4/21/07, open-graphics <[EMAIL PROTECTED]> wrote:
are any of these GPU's really that much faster than a single Intel
or AMD core at these math functions?
Most definitely. Check out GPGPU and see how GPUs are being used to
make 'supercomputers'. GPUs are very specialized to their function,
and that specialization means that they've vastly more efficient at a
certain kind of processing.
Someone else can give an accurate account of the performance boost,
but pages I'm brining up are suggesting that one GPU is roughly 10x
faster than a GPU at the same task, when the algorithm is adapted
properly.
If you design a custom board to do this kid of 3D, it's always going to
be a generation behind the top of the line, and it's going to be more
expensive to produce because of the niche popularity. Less power for
more money is never a good way to do anything.
Our GPU will be generations behind the other GPUs but still faster
than CPUs at graphics.
That being said, I'm encouraging some of the others on the list to
design a CPU or DSP-based graphics card, at least as an academic
exercise.
Please note that this discussion of using a CPU core for graphics is a
really old one and has been brought up and torn down many times. You
should have a look at the archives.
The thing that is most badly needed is a pci(e) card for experimenters
with the pci firmware included so that other things can be *soldered* or
*connected* by experimenters, such as extra CPUs or daughter boards,
without having to get bogged down in pci core problems and cost.
That board alone will make more money years ahead of any graphics
functions. It will also give real experimenters real things (including
graphics) to try instead of dreaming about it.
The economies of scale that follow can be leveraged for all users in
gaining hardware experience and aid funding of the eventual graphics
functions. Atleast the card could be immediately useful as a dumb
frame buffer configuration for users to start adding graphics functions
to incrementally.
3D openGL and games garbage is at the bottom of the list of most
engineering users of an item like this, and yet have the most money
and are willing to spend it for something useful. You won't get money
from high-end gamers for a while yet. I'd focus on industrial users,
engineers, and hackers.
--
Russell Shaw, B.Eng, M.Eng(Research)
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
