On Aug 24, 1:04 pm, DrQ <
redr...@yahoo.com> wrote:
> What does GPU stand for? Graphical Processor Unit? Wrong!
>
> **GPUs are the new vector CPUs**
>
> Oh! And don't forget Intel's Ct. Anybody know how that stacks up
> against CUDA?
There a discussion of using CUDA for parallel database search in this
month's USENIX login:
Programming Video Cards for Database Applications by Tim Kaldewey
http://www.usenix.org/publications/login/2009-08/index.html
alternate version:
Parallel Search On Video Cards
Tim Kaldewey, Jeff Hagen, Andrea Di Blas, Eric Sedlar
Oracle Server Technologies - Special Projects
http://www.usenix.org/event/hotpar09/tech/full_papers/kaldeway/kaldeway_html/
I think the story is the same as always, if you have a highly parallel
problem, GPUs are a huge win for larger scales of data. Applying gobs
of commodity hardware to approach the performance of supercomputers at
a fraction of the price is always appealing (think Hadoop).
However, it is not a magic bullet, or magic feather, and there will be
application of GPUs pursued to utter failure; they will make great war
stories. Knowing when to apply them will mean all the difference
between success and failure.