On Mon, 18 Oct 2004, Niall Dalton wrote:
The proceedings and talks from the last ACM workshop on this topic
makes for interesting reading.
> http://www.cs.unc.edu/Events/Conferences/GP2/program.shtml
I could be mistaken, but it seems to me that in all direct performance
comparisons between GPU and CPU, vectorized CPU programs are
conspiciously
absent.
If you do go to the length to write hardware specific programs for the
GPU
to gain, say, a factor of four speed improvement, then you can as well
write hardware specific AltiVec ('velocity engine') code to gain pretty
much equivalent amounts of performance.
