Re: SuSE 10.0 boot problems with CPU changed


Aragorn wrote:

No... ;-) Hardware is all kernelspace stuff in GNU/Linux, and a CPU is
essential to get the system to boot in the first place. ;-)

OK, thanks.

I'm really trapped between the devil & the deep blue sea. Can't afford
stock performance of better hardware but really need the FLOPS.

Is there no possibility of using some of the University's CPU cycles for
your computations? Some Universities allow their students to make use of
their big iron for project-related stuff... ;-)

Yeah. My university has a parallel cluster which I have access to.
About 30 SGI Altix, 200 dual core Xeons, 130 quad core Xeons (just
installed) and 1600 more nodes planned soon.

My problem is their queueing system. The nature of my work requires
hundreds to thousands of short-ish runs, rather than one behemoth run.
As such I have to queue hundreds to thousands of times, which bumps up
the "apparent" processing time a lot.

Consider that a typical job of mine that takes about 5 minutes to run
on a P4 1.7GHz CPU. Sometimes queue times are 45 minutes...

Another option would be to buy up some older second-hand PC's - they should
come in cheap - and build your own Beowulf cluster. /That/ should give you
some flops... ;-)

The problem with Beowulfing is that my source codes have not been
adapted to run in asynchronous parallel mode yet. Which means that slow
CPUs will bottleneck the whole run.

So far I have found the most cost effective solution to be an
overclocked Intel D 805 CPU. I overclocked it from 2x2.66GHz to
2x3.33GHz and it has been running at ~2x100% CPU utilisation for almost
a solid week with no problems. I am using DDR2-667 RAM.

People have managed to overclock this monster to 2x4.0GHz, but IMVHO
the premium you need to pay for quality DDR2-800 RAM/mobo + water
cooling offsets the returns you get on a faster CPU. For the price of
one 2x4.0GHz system I could probably get two 2x3.33GHz systems.

.