We are hopefully getting a high performance cluster at my work soon (which I will hopefully be administering.) It should comprise about 100 Opteron CPUs. Not as cool as the above, of course, but nice.

lol, still very impressive. What are you going to be using that computer for?

If we get this kit, it will be being used to do some mathematical computations. We have several PhD students here, also Professors and research assistants that want to do some calculations that even on currently fairly powerful PCs just take far too long (I am talking one experiment taking over one month!) We are looking at ways to calculate these things more quickly, without getting a true supercomputer, which is beyond our budget.

there is much more to it than that, of course. certain types of tasks compute better and faster on parallel setups like the kind of cluster that we want, and tailoring it to our needs is almost a full time job.

We are hopefully getting something like this:

http://www.cray.com/products/systems/hpccluster/

Sweet.. what kind of job do you have/what are you caluclating? Interesting

I am a computer technician / network administrator at a university.

Essentially, the kit is used for number crunching (computations with large matrices, etc).

and, doesn't the amd processor

do that (number crunching) ..............best of all?

JSR....those are very big mambo jambo calculations. Not just simple 1 +1 ..ehhehe .

young

you may be able to beat up on the x man, but, don't mess with a sr, ......jr...... an opteron processor is, an amd part....and, when the hammer arrives, this next month, ........that too, will also be an amd part......and, my question to phant, is simply this........that the amd component, pound for pound..........even in the 1+1 processors (as this was manifested to me by a washington state math student, whom had similar aspirations in math calc, as does phant), were, and still are, better for number crunching huge applications, and that is the reason for their worth, and, continued popularity.

Re: young

I think i got your previous post wrong. Try to write something in a more organised way or else you might end up confusing alot of people.


P.S Pls stay away from drugs.

yes, you did

could you keep your crib, .....in a more organized manner
p.s. stay off the nipple :grinthumb

Re: Re: young

True.

yes, i have

always...............known this :grinthumb but, i know everyone ......hehe...............accepts me :darth: ........warts and all ;P so, phant, back to my question.......does the amd processors prove to be the best at number crunching? i have always believed so, but, it would be nice to retrieve some corroboration from you

We don't actually have it, yet. We may still not get the budget.

If we do, in about 6 months, I will be able to answer your question.

ok

thx,....... i believe you have made the right choice, for the direction you are headed.

Remember, of course, that this is a cluster of machines, not one singular monster of a computer...

But when you invoke certain software, the cluster ACTS AS ONE MACHINE. At least, that's the theory. I still have a lot of work to do in this area, as it invovles writing optimised code and so forth.

10/4

no,...... i copy that. it obviously is the path being taken by many.........the leapfrog to new architecture should be a boon

Basically, clustering (which isn't that new) is increasingly being seen as a cheap alternative to supercomputing. At least, that's the theory.

A secondary, related theory is that 64-bit computing on clusters will help to further augment this.

here's a link

http://www.aspsys.com/about/news/press/default.aspx/6.aspx

That not a bad link. I just read that and it was interesting, albeit brief.

With high speed myrinet, that ASPEN system is probably quite a beast.

With clusters, its not just the speed of the individual nodes, or how many you have, but also how fast the nodes can talk to each other which is the real factor. Network speed between nodes can be a real bottleneck, since the whole technology is based on the notion of "passing messages." Indeed, the implimentation I will initially be looking at (if we get the money) will be MPI, which is "Message Passing Interface." There is an MPI optimised compiler, and MPI runtime tools.