comparing CPU's

[ted leslie]

Yes this seems very true, about how CPU speed seems to becoming less and
less important.

Just got a New dual Opteron TYAN MOBO
and outfit with dual 2.4Ghz cpu's
before i had a dual athlon MP 2800, which is only a 2.0Ghz clock,
and the new opties have bigger L1/L2,

when i first used the new dual opty it really wasnt all that quickier,
in fact i hardly noticed a difference,

but when i put 9GB ram in it, and a 10K raptor
and a Nvidia 7900 vid card
and those additions made way more of a difference then the processor
upgrade.

I got the 9GB of ram because I am planning to install a vmware install 
of a linux into a RAM disk (i.e. whole OS in RAM) just to see what it
would be like.

Looks like the days of getting more (real app) power from cpu's is
finally over, and RAM, video card and HD seem to impact much more.
With single CPU's become extinct and dual's being the norm, and in a
year quadies being the norm, linux and and apps on linux have to start
being smart about concurrency other wise we might lose some ground to
Bill.

At this time, I have to think the disappointment is coming primarily
from drive seeks, moores law has been fairly accurate up until a year or
(and technically moores law has nothing to do with HD's, but it just
that whole "you like to see your whole 'power' double every 18 months'"
rule) so ago, but when it comes to real drive seek times, i think they
have only about doubled in the last 8 years, and when OpenOffice
references about 100-150+ *.so files on a full boot up and use, thats
probably upwards of 2 seconds just on drive seeking and reading.

I am planning (when i get some hack time) to put a cluster FS on my box,
and have one of the clusters in RAM and the other on disk, 
the secondary will be the disk one, and I will keep all system libs, etc
in RAM on that ram/cluster primary, and when i update a package, 
it will propagate to the secondary, and then be there for the reboot,
when it will reset up the cluster and ram disk. 

This is a better use of ram then doing everything in a RAM vmware
machine. Plus you can't use compiz through vmware, and once you have
used compiz, its not like you can go back to a old fashion linux desktop
environment, even if compiz is 50% eye candy. Nice thing about compiz is
you can tell your "windows' using friends" (or Mac users) that your
where they will be "on the desktop" on vista (or Jag), in about a year,
except I'm (you're) there TODAY! :) :) [this is the best linux gloating
ammo thats been provided to us in years!]


-tl



On Sun, 2006-08-27 at 23:48 -0400, Christopher Browne wrote:
> On 8/27/06, Dave Cramer <davec-zxk95TxsVYDyHADnj0MGvQC/G2K4zDHf at public.gmane.org> wrote:
> > Now that Intel finally has something to compare to the Opteron, I'm
> > wondering how to compare them.
> 
> It's not evident that Intel has done anything to address the memory
> bandwidth differences, which has been where AMD-64 has been
> particularly better than EM64-T...  Even if the latest Intel CPUs are
> faster, if they haven't addressed the memory bus issues, they don't
> likely have parity let alone superiority.
> 
> After all, the cases where enormous amounts of CPU power can get used
> tend to be few and far between; it's commonly I/O and memory bandwidth
> that are the bottlenecks.  And the cases where I see CPU used up are
> cases where it's busy throwing data around in memory, in which case it
> is memory bandwidth that's the bottleneck...

--
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