[SWLUG] how much more CPU performance can one get these days?

[Huw Lynes]

On Fri, 2009-09-18 at 11:17 +0100, Mark Summerfield wrote:
> On 2009-09-18, Huw Lynes wrote:
> > On Fri, 2009-09-18 at 08:08 +0100, Mark Summerfield wrote:
> > > Hi,
> > >
> > > Short version:
> > >     I have Linux 2.6 2GB RAM Dual Core Pentium D 2.8GHz
> > >     What would I need to get 5x the CPU performance?
> > >
> > >
> > > Long version:
> > >
> > >     I sometimes have to run a program that starts by reading 5-50MB of
> > >     data, then spends 12-15 minutes processing (no I/O, no graphics,
> > >     pure CPU), then writes out a single 5-100MB output file.
> > 
> > I'm assuming that this code is doing floating-point maths rather than
> > integer maths? If so you will see a big speed-up on core2 or above.
> 
> Yes it is. What does "core2" mean & how would I know what would give me
> that? (I stopped following processor chips way back when the "286" was a
> big deal...)


Intel's current stable architecture. Core2 Duo being the equivalent of
the chip in your machine. i7 is the new shiny chip, but it comes with a
shiny price-tag while core2 is currently nose-diving through the
pricelists.

Your current chip is based on the old netburst architecture. The core
architecture has notable advantages such as running two floating-point
operations on each FPU per clock cycle. And of course all the modern
chips have a higher memory bus speed which can't hurt.

That being said if the code itself is badly written it doesn't matter
how much hardware is thrown at it, it won't go faster. Which is why I
recommend testing it rather that making a speculative hardware purchase.

-- 
Huw Lynes                       | Advanced Research Computing
HEC Sysadmin                    | Cardiff University
                                | Redwood Building, 
Tel: +44 (0) 29208 70626        | King Edward VII Avenue, CF10 3NB