Hello, been a while, dual CPU mobos

[The Edge of the Ice]

On Wed, 08 Sep 2004 20:01:06 -0400, Andrew Hammond
<ahammond-swQf4SbcV9C7WVzo/KQ3Mw at public.gmane.org> wrote:
> Profiling your existing system using vmstat, as Ralph suggested earlier
> in the thread, and posting the results would provide at least something
> to work with.

Someone mentioned early on regarding OLD SMP systems versus a NEW uniprocessor,
which is of course entirely specious.  Something that vmstat will NOT
tell you (unless
it does much more than meets the eye) is how much memory and cache latency
are affecting your program's performance.  My G4 800MHz (7451, PC-133,
64/256/1024k L1/2/3) PowerBook beat my friend's G4 867MHz (7455, PC-266,
64/256/0k L1/2/3) PowerBook by a good 5-10% on a neural nets homework
assignment we had.  I'm pretty sure that the entire program itself,
and probably a good
chunk of the dataset fit inside my L3 cache, leading directly to the
lion's share of that
bonus.

Not that you have the choice in cache these days like you used to. 
But if you're doing
cryptanalysis work with relatively small datasets and small codesize,
I'd place bets
on bigger caches which are able to entirely contain your work having a
positive impact
on speed.

I also hear that the newer Pentium (Pentium4, at least) chips have
moved away from
doing bit operations in hardware, and instead perform them (at a cost)
via microcode.
That might be why my Pentium4-M 1.8GHz laptop gets is ass kicked by my Duron 1G
when it comes to compiling.

So, if you're looking for bang-for-buck for a very specific set of
applications, I'd suggest
that you find people or stores who are willing to let you run some
test programs, and see
exactly how well the hardware performs in the REAL world, with YOUR workload.

BTW, I'll have to remember that one doesn't need a cluster to do nuclear weapons
research.  ;)

-- 
taa
/*eof*/
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml