Date: Fri, 25 Feb 2005 09:52:49 -0900
Reply-To: Nels Tomlinson <nels_tomlinson@labor.state.ak.us>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Nels Tomlinson <nels_tomlinson@labor.state.ak.us>
Subject: Re: Best PC for SPSS with large datasets
In-Reply-To: <421F0818.8080702@york.ac.uk>
Content-type: text/plain; charset=us-ascii; format=flowed
On the dual processor question, you need to ask whether your OS and
applications are going to be able to make use of the second CPU. Does
Windows-whatever use extra CPUs? I know that Linux will, and I think
that FreeBSD will (and OpenBSD won't), but you may have to pay extra for
a server version of Windows if you want your second CPU to be more than
an expensive space heater.
Then there's the application itself. Does SPSS make use of multiple
CPUs? If it isn't written to be threaded, it probably doesn't. If the
OS will use multiple CPUs but the application won't, at least the
application can have one CPU all to itself, and the machine and the
application may seem a bit more responsive. You'll be able to read your
email and kill time without slowing down your program.
As for RAM, I'd say that it's still cheap enough that you should have as
much as the machine will hold. On this machine I'm using, that's
2GBytes. Of course you want the fastest RAM available and memory
bandwidth (FSB speed) is often more important than CPU speed.
I don't know that dual processors limit FSB speed. That sounds like a
limitation of either one manufacturer's chipset, or one manufacturer's
CPUs. If you've only been looking at Intel boxes, do also consider
AMD's Athlons. Right now, AMD seems to be a bit ahead of Intel for
floating point math, but also look at Intel if you haven't yet; they
aren't out of the running yet. If you're willing to give up Windows,
also look at Macs and IBM PowerPCs. They have 64 bit CPUs with FAST
vector math, and dual (or more!) CPUs if your software can use it.
If you're going to run Windows, I think that two hard drives would be a
good plan, IF you can have them on separate IDE busses. That should be
do-able on most PCs. 7,500 RPM hard drives of moderate size are getting
affordable. If you have huge data files, put them in a partition at the
outside edge of the drive, where the sectors are longer and the drive
won't have to step quite so often (on any OS). Remember that the
fastest drive is going to be around 1,000 times slower than the slowest
RAM, so you want to avoid using the drives when you can!
If you're buying a Dell or an HP, you have to take what they give you,
and it may or may not work for you. There's a good chance that you can
make a better machine, with higher quality parts, for a similar cost, if
you buy a suitable motherboard and parts, and assemble your own. If
this is a work machine, that may not be possible, but for a home
machine, you can get a proper workstation at a modest savings that way.
I probably wouldn't do this with a Windows machine: installing Windows
is difficult, and if it doesn't work, you're dead in the water.
Hope this helps,
Nels
Jeremy Miles wrote:
> Hello All,
>
> I've seen this question come up in the past, but not for a long time,
> and as things move on, I thought I would see if anyone had any thoughts
> on it. If one were buying a new PC to run SPSS, with some pretty large
> datasets, what criteria would you go for.
>
> Obviously lots of RAM, lots of disk, fast processor. But does anyone
> have thoughts / experience of dual processor systems? I have been told
> that dual processors limit the speed of the FSB to 533 MHz, but that a
> single can be up to 800 MHz, and that this trade-off eliminates the
> benefit of the two processors.
>
> How much RAM is worth getting? Windows/SPSS overhead + max file size?
> WIndows overhead + max file size * 2, where max file size is the biggest
> file size you work with in SPSS.
>
> Are two hard disks worth while, to keep Windows on one, and programs /
> data on the other?
>
> Any thoughts appreciated. If people want to reply directly to me, I
> will summarise to the list.
>
> Thanks,
>
> Jeremy
>
>
>
>
>
> --
> Jeremy Miles
> mailto:jnvm1@york.ac.uk http://www-users.york.ac.uk/~jnvm1/
> Dept of Health Sciences (Area 4), University of York, York, YO10 5DD
> Phone: 01904 321375 Mobile: 07941 228018 Fax 01904 321320
>
| 
