Computer building opinions

[jeffs]

Why make it yourself. All you need is money

http://www.oracle.com/us/products/se...er-077287.html

[Brian d marge]

Btw I did forget to say

I use linux 64 bit debian and I feel like its going to be , the I7 quad for a start , then followed by another of the same as the prices will have dropped by then

That ( While a long way of , ) will be enough to keep me happy till end of next year

and lots and lots of Rams

Canterbury one of course

Stephen

[Grasshopperus]

Before you go off and start looking at any hardware you should work to understand the problem.

Some equations do not benefit from more execution cores, for example; the Fibonacci sequence involves starting a sequence with a "0,1" and then figuring out each subsequent number as a sum of the two previous numbers eg. Fn = F(n-1) + F(n-2). In this situation you have to complete an addition before moving onto the next number, you can't start in an arbitrary location so the second core wouldn't be able to help at all.

For memory, are you actually maxing it out at the moment? Most calculations are CPU intensive as opposed to RAM intensive.

As for using a PS3, if your equation deals with floating point numbers, matrices or other types of data that can be optimized for "Single Instruction Multiple Data" instructions then they can be useful. But they're not the supercomputers in a box solution that popular media makes them out to be.

I've no idea what type of calculation you're doing but justly blindly spending money and hoping it'll make your problems go away could be a bit of waste.

[Gremlin]

It'll be damn fun doing it tho, until you see the invoice.

[p.dath]

Before you go off and start looking at any hardware you should work to understand the problem.

Some equations do not benefit from more execution cores, for example; the Fibonacci sequence involves starting a sequence with a "0,1" and then figuring out each subsequent number as a sum of the two previous numbers eg. Fn = F(n-1) + F(n-2). In this situation you have to complete an addition before moving onto the next number, you can't start in an arbitrary location so the second core wouldn't be able to help at all.

May I prefer you to the Intel parallel processing language Cilk, and this document describing how to calculate the Fibonacci sequence using parallel programming.
http://supertech.csail.mit.edu/cilk/manual-5.4.6.pdf

Often it is a matter of thinking differently to be able to work in parallel.

[Brian d marge]

At the moment it looks like a "render farm" I was thinking about it last night ,

Custom made chassis ( cant be any harder than motorbike bits) then mother boards and chips and 2 gig per core

Thats quite cheap , infact I could build each node for less than 10 000 yen ( expanding as I get better

The master computer , will be the bloomfield chip with as much ram as it will take

I need to research this a lot more but thanks to u lot a few more avenues appeared
ta!

Stephen

[Juzz976]

Whats with the obsession with having all this RAM, do you people even look at how much is actually being used.

2GB per core, wtf thats retarded. So glad I'm not in the IT industry anymore, been taken over by over excitable numpties.

As you were

[Latte]

Some modern apps are very resource inefficient , I struggle with 2gb physical at work (VMWare console, 2 .net apps and a java app and it's all over).

[Brian d marge]

Whats with the obsession with having all this RAM, do you people even look at how much is actually being used.

2GB per core, wtf thats retarded. So glad I'm not in the IT industry anymore, been taken over by over excitable numpties.

As you were

I know how you feel

I have to listen to you when you bring your bike in for a service ,,,,

I was advised by someone I trust that around 2 gig per core is about right for Openfoam

soo,,,,, Im going with this ,,,

Dont feel bad if I put music on when your explaining about your pride and joy ,,,

Stephen

oh ps , why is it retarded in your Humble opinion

PPs , yes from memory over (70 % ,,, I have this wonderful system monitor in the tray , both cores near 90% ,,,,,Im running another solver in a few min

would sir like a screen shot??

[Mental Trousers]

Here is what Open foam does

From my home page

http://477-racing.webs.com/apps/videos/

Don't show me that!!!! Now I want to install OpenFoam and model the intake and exhaust on my race bike

[Grasshopperus]

May I prefer you to the Intel parallel processing language Cilk, and this document describing how to calculate the Fibonacci sequence using parallel programming.
http://supertech.csail.mit.edu/cilk/manual-5.4.6.pdf

Often it is a matter of thinking differently to be able to work in parallel.

Ha, got you you bastard

That example is just a lightweight demonstration of CLIK's message passing interface.

From the doco,
"In the Fibonacci example, a sync statement is required before the statement return (x+y), to avoid the anomaly that would occur if x and y were summed before each had been computed."

So this is simply making a parallel cluster work in a serial fashion by farming out each sequence number calculation to the cluster, getting the response and then farming out a request for the next number. In this case the network overhead for the sync would defeat the point. There is no true parallel processing here.

[Brian d marge]

Don't show me that!!!! Now I want to install OpenFoam and model the intake and exhaust on my race bike

its free at CAElinux

Meshing is the hard part.. but if you keep it simple and get a water tight , non manifold part it will run soothly

and yes , that what I do ,,,,,

( my next project is flow through an suspension mid valve , ,,,,tested by using shapeways 3d printer ............)

Stephen

[p.dath]

Whats with the obsession with having all this RAM, do you people even look at how much is actually being used.

He is probably working with decent size data sets, and holding the entire data set in RAM makes the processing much faster.

Ha, got you you bastard ...There is no true parallel processing here.

Doh!

[p.dath]

Have you considered simply renting cloud resource, like from Amazon and the like? You can pay by the hour of CPU time used.

[Brian d marge]

running a simple routine ( or whatever you call it now )

course grid, a few million cells and its

CPU 1 100% Cpu2 40 % Ram Just short of one Gig ( 800 ) and will be complete in 40 min ( and produce 5 gig of data)

All this will give is a rough idea of airflow over a VW beetle ( no reason)

and when I use parafoam to view the results , ( another area thats intensive ) every pass on the above data set takes ( up to ) 2 seconds per iteration ( this one has 1000 iterations )

So there u have it


Stephen

Beowulf cluster>>>???