Re: [OxLUG] What would make the biggest difference to my pro…

Top Page
Attachments:
Message as email
+ (text/plain)
Delete this message
Reply to this message
Author: Chris Wareham
Date:  
To: Oxfordshire Linux User Group Discussion List
Subject: Re: [OxLUG] What would make the biggest difference to my processing speed?
On 08/08/11 10:30, John Winters wrote:
> A request for hardware advice if I may.
>
> I currently use a box running Debian Squeeze which has a Gigabyte
> motherboard (GA-G41MT-D3) with a Pentium dual-core processor (E5800
> 3.2GHz). I use it for some program development and occasionally some
> video processing.
>
> One particular job which I do (processing a big heap of data for my
> current programming project) takes about 8 minutes of hammering one core
> of the processor, whilst processing a video can take well over half an
> hour.
>
> It's not constrained for RAM - 4G and never swaps.
>
> I'm wondering about changing the CPU - it seems the board supports a
> wide range of faster processors:
>
> http://www.gigabyte.com/support-downloads/cpu-support-popup.aspx?pid=3476
>
> but there are several different variables which might make a difference.
>
> The current processor has 2M of cache, and an FSB at 800. Alternative
> Core-2 Duo and Core-2 Quad processors offer variously 3M, 4M, 6M or 12M
> of L2 cache, and an FSB at 1066 or 1333, and on the whole they actually
> have slower processor clock speeds.
>
> I don't think extra cores are going to make much difference for my
> mainly single-thread purposes, so of the other variables, which is
> likely to make the most difference to number crunching? I don't want to
> have to spend £300-400 on a top-of-the range CPU. I'm thinking more in
> terms of £150, and at that price you can have some but not all of the
> above enhancements.
>
> Grateful for any advice.
>
> Cheers,
> John
>


I'd strongly recommend going for the faster FSB. On a number of
occasions when benchmarking different CPUs (AMD Opteron versus Intel
Xeon for example), a slower clocked chip with a faster or wider memory
bus will outperform a faster clocked one with a slower or narrower bus.
The bottleneck for such highly clocked CPUs can be getting data in or
out of the processor quickly enough to take advantage of the clock
speed. With multiple cores and workloads that can be processed in a
highly threaded manner - such as transcoding video - this can make a
considerable difference to overall utilisation of the CPU.

I've not compared them for a while, but if the current crop of Intel
chips still use fundamentally the same FSB technology as they did a few
years back, then an AMD processor with HyperTransport should be better
for parallel processing needs.

Chris