10 NPE P4 3ghz Cluster useful?

[gradstudent101]

Howdy!

I'm a grad student working in a lab that has a stack of P4 desktops, about 10 altogether that run at 3ghz with 1gig ram.  They're just standing around and I was thinking of making a cluster computer from them.  Does anyone know if this is a waste of time?  I can't seem to find performance specs for similar setups online.

Thanks!

[Calum]

Depends what you plan on doing with the cluster.  Clusters aren't often implemented outside of certain rather specialised situations due to the nature of cluster computing.
If they're spare and you have plans, then sure they could be useful - but if you're just looking for something to do with them, for the power and heat they're most likely not worth running and you would be better off selling them all to buy one or two modern PCs.

P.S. welcome to the forums!

[gradstudent101]

Hi Calum, thanks for the reply and the warm welcome.  I do physics and optimization problems come up very often.  I was stuck with an 8th order equation last year, if I solved it I could have published, but alas, Mathematica ran several days without giving a solution.  I was interested in the potential processing power of such a cluster and whether it would beat out an i9 or something else that's new and shnazzy.  It wouldn't be on all the time probably, just for calculations.  Mathematica is a cluster-friendly program btw.

Cheers!

[Calum]

I'm no expert on clusters, hence my somewhat vague reply.
If Mathematica is something that can take great advantage of clusters, on your cluster it could perform theoretically as well if it was running as a 10-core P4 3GHz (if such a beast existed).  In terms of pure computing power, they would be roughly between an i5 750 and an i7 860.  Yes - "an", meaning just one...amazing how far we've come.  My advice would be to see how well the cluster does at solving problems as you describe, but my gut feeling still tells me you'd be better selling them all to purchase a modern machine - if they all went for $150, that's $1500, which could quite possibly net you two i7 machines (assuming you want these for just number crunching, no fancy graphics cards/storage capabilities etc) - a vast increase in computing power, less power draw, basically win/win.
Just my £0.02, anyway.

[gradstudent101]

Thanks again for the change.   My research advisor just told me that those machines used to be a cluster but it was disassembled some time ago, so I've been given the ok to restore it.  In fact parallel machines speed up computing time in a non-linear way, ie having two machines instead of one increases the efficiency three-fold, so it is reasonable to expect them to outperform today's models.  I doubt I could sell them for much anyway and it seems like a fun project so wish me luck!

[patio]

Best of Luck to you and Welcome Aboard ! !

[Calum]

In fact parallel machines speed up computing time in a non-linear way, ie having two machines instead of one increases the efficiency three-fold, so it is reasonable to expect them to outperform today's models.

Interesting.
Well, that's great then - happy crunching, and good luck!

[Salmon Trout]

You can read about Beowulf clusters on Slashdot.

[rthompson80819]

When your not doing calculations, you might check out folding@home for a use for the computers.

http://www.computerhope.com/forum/index.php/topic,77650.0.html