os that rather uses the gpu?
Robert Myers
rbmyersusa at gmail.com
Sat Jul 17 02:19:53 UTC 2010
On Fri, Jul 16, 2010 at 9:49 PM, Genes MailLists <lists at sapience.com> wrote:
> On 07/16/2010 08:07 PM, Robert Myers wrote:
>
> We ett that this stuff may not help your problem domain - don't
> assume all problems look like the one(s) you are interested in.
>
> This stuff is useful - I have seen it as have many others ... it is
> not a panacea ... stop harping please ... and try understand the useful
> side as well as the limitations .. it has limitations - as do people.
>
If you wanted this discussion to end, why are you continuing it?
Gordon Bell asked (about huge computers) in a presentation now 12 years old:
"Is it worth it?"
If you're not going to get new science out of a huge new computer, then why
spend the money?
I have one answer to where there is possibly new science, but we never get
the computer to explore it.
"My" problem domain (fluid mechanics) is a big driver behind the huge
expenditures on "supercomputers," which as I have pointed out ad nauseum,
are far from super in some really important ways.
The GPU may be a very handy tool for certain kinds of computations, but it
almost inherently replicates a design corner-cutting that that is endemic in
supercomputing right now (low bytes per flop).
If you want to say, "GPGPU's will help *my* problem, so I want one," I have
no answer for that.
If you propose that the GPU is an ideal tool for university level teaching
and research that requires hefty computing, as the poster I responded to
did, then I have some serious objections based on good science.
I presented my position, I defended my reputation and my position against
wild and ignorant slurs, and you have now joined a chorus that wants to
comment on my personality, rather than on anything technical. Perhaps you
are the one who should stop harping?
Robert.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.fedoraproject.org/pipermail/users/attachments/20100716/eaedd070/attachment.html
More information about the users
mailing list