quad cores vs not.

GCC Consulting gccconsulting at comcast.net
Fri Mar 21 06:46:20 PDT 2008



> -----Original Message-----
> From: filepro-list-
> bounces+gccconsulting=comcast.net at lists.celestial.com [mailto:filepro-
> list-bounces+gccconsulting=comcast.net at lists.celestial.com] On Behalf
> Of Fairlight
> Sent: Thursday, March 20, 2008 7:35 PM
> To: filePro Mailing List
> Subject: OT: Re: quad cores vs not.
> 
> Walter brought up quad cores in another thread...
> 
> The reality is that, available or not, quad cores are almost useless
> outside of a strict server environment.  It's not like they're rubbish,
> but
> there are -very- few things that will utilise all the cores on a
> desktop
> system:
> 
> * CAD - Some probably will
> 
> * Games - Mixed.  Many single-core games (even current releases--Tabula
>   Rasa only used one of two cores), some dual core.  Most won't make
> use
>   of more than two.
> 
> * Applications - Depends how threaded an application is.  Varies.
> 
> * Video Editing/Transcoding - One of the most CPU-intensive tasks
> anyone
>   will be doing on a desktop, editors are often multithreaded and make
> use
>   of multiple cores, but the -important- part, the codec, is often a
> single
>   core application, non-multithreaded.  The only multithreaded codec I
> can
>   confirm is WM9.  Even XViD is singlethreaded.  The codec is honestly
>   where you want firepower to count.
> 
> * BOINC infrastructure based applications - Will make use of as much as
> you
>   can toss at them.  Great if you spend money to support seti at home and
> all
>   the other @home projects, but otherwise not so impressive.
> 
> * Graphics - SOME graphics renderers will make use of multiple cores.
> Some
>   do it in a limited fashion.  Also highly processor intensive, yet
>   programs like Bryce don't even make use of a second core.
> 
> The reality is that you're better off with something like a Wolfdale
> dual
> core than a quad core at this point in time, and probably for several
> years
> (3-5 would be my guess).  You're better off throwing 3GHz dual core at
> something where at least most of it will be used, than you are throwing
> 4 cores or more at a lower clocking at things that won't use it.
> Unless
> you multitask, it's a LOT of expensive wasted firepower.  You can
> actually
> outperform a quad core with a dual core in many cases.  I considered
> going
> quad (Q6600) and was advised that the E8400 dual was a better bet at
> this
> point in time, especially for my uses.
> 
> I'd recommend the Intel E8400 3GHz Wolfdale in a heartbeat.  They're
> currently in short supply due to popularity, but if you can get one,
> they're -great-.  There's one Wolfdale that's 0.166 GHz faster for like
> $120 more--totally not worth it.
> 
> I also recommend going for chips with reduced die sizes.  That was one
> advantage of the Wolfdale series, was the reduced die size.  That
> translates to a LOT cooler chip and less power consumption.  I couldn't
> give much of a damn about power consumption, but heat == death, so
> cooler
> is better.  This thing runs ridiculously cool.  In fact, the memory
> controller hub on the mainboard runs hotter than the chip itself, as
> does
> the I/O controller.
> 
> I was told that Intel almost religiously releases every 11 months.
> They'll
> release one line with increased performance, then 11 months later
> they'll
> release the same line with -slight- improvements but mostly just a
> reduced
> die size.  Those are the ones to get.
> 
> mark->

Wish list:

Love to see filePro able to make use of the multi-core processors.  This is
especially true of the windows version as one can't have background
operations as the *nix version have. Even better, moving some of these
background processes to other processors on a nix system.

Or, how about a 64bit version of FP running on multiple cores?  Now that's
an interesting question, what additional features could be added to a 64bit
version of fp?

Richard Kreiss
GCC Consulting
rkreiss at gccconsulting.net
  







More information about the Filepro-list mailing list