Mini-ITX 12-node cluster

8 posts / 0 new
Last post
Dr. Webster's picture
Offline
Last seen: 13 hours 20 min ago
Joined: Dec 19 2003 - 17:34
Posts: 1747
Mini-ITX 12-node cluster

http://www.mini-itx.com/projects/cluster/

Looks way sweet.

catmistake's picture
Offline
Last seen: 2 years 4 months ago
Joined: Dec 20 2003 - 10:38
Posts: 1100
saw this awile ago...

"achieving a performance of around 3.6 GFLP"

Doesn't it seem grossly underpowered?

I've clocked my 1GHz G4 Powerbook at 4 gigaflops...

Why isn't that cluster a screamer?

I'd like to see a nano-itx cluser, filling up an ANS chassis with as many nodes that will fit (lots and lots).

http://news.bbc.co.uk/2/hi/technology/3532706.stm

The new nasa cluster should be interesting, but why, oh why, do they have to fall in with the penguinistas?

this one, too

http://arrakis.ncsa.uiuc.edu/ps2/cluster.php

Dr. Webster's picture
Offline
Last seen: 13 hours 20 min ago
Joined: Dec 19 2003 - 17:34
Posts: 1747
The reason that thing is so s

The reason that thing is so slow is because they're all VIA C3 series processors. Very low-power chips, best used for basic computing. But at such low cost, making a cluster out of them can be appealing. I doubt he spent much more than $1500 on that cluster.

catmistake's picture
Offline
Last seen: 2 years 4 months ago
Joined: Dec 20 2003 - 10:38
Posts: 1100
and low Wattage

only 120 Watts at idle? not bad

odd...
"looks to be equivalent to at least 4 (maybe 6) 2.4Ghz Pentium IV boxes in parallel on a similar network "

but... this just seems wrong... 4 X 2.4GHz Pentium IVs, as crappy as I think they are, would have to do better than 3.6 gigaflops, right?

Dr. Webster's picture
Offline
Last seen: 13 hours 20 min ago
Joined: Dec 19 2003 - 17:34
Posts: 1747
A gigaflop is a lot more proc

A gigaflop is a lot more processing power than you think. I honestly doubt that you got 3 Gfp on your PowerBook..the highest end Power Macs don't get much better than 1 Gfp. 3.6 Gfps for four 2.4GHz P4's sounds about right to me.

catmistake's picture
Offline
Last seen: 2 years 4 months ago
Joined: Dec 20 2003 - 10:38
Posts: 1100
AltiVec Fractal Carbon

Hey, I'm no expert, I got this piece of software, it said so...

its not always above 4000 megaflops, usually around 3600, but sometimes its up there.

Know of any simple, say, command line benchmarking utilities?

performaman's picture
Offline
Last seen: 17 years 8 months ago
Joined: Dec 20 2003 - 10:38
Posts: 209
Darn, no games!

Only problem is that there aren't many apps for clustered computing that the average (Applefritter) user would use. What I mean is that there are no cluster-based games. Of course, this is because the market is so small for these games. What would be cool is if there was an intermediate program that would take instructions from a program not designed for clusters and divide the instructions among the diffrent nodes. Then, the program would think that it was running on one really fast processor while in reality it would be running on many processors. You'd probably have to write an entirely diffrent OS, or do some serious software hacking on a current one to make this work, but it would be cool.

Jon
Jon's picture
Offline
Last seen: 12 years 10 months ago
Joined: Dec 20 2003 - 10:38
Posts: 2804
The problem is a compile tim

The problem is a compile time issue. Taking it on at run time would be slower. When writing apps that use clustering or parallel processing you write the code and specify chunks that can be done in parallel. Convincing a run time program to make the decisions of what needs to be done in sequence andd what needs to be done in parallel would require much analysis of the code by the program, essentially training it to do what should have been done when the program is compiled. Modern proccessors and compilers already do some of this, but it is all tuned for single processors.

A decent modern OS can run multiple threads on different CPUs to get better performance, but each thread will only run as fast as the CPU that it is on. If the texturing code or what ever is running on one thread, that's the single CPU bottle neck again, 'cause everything else has to wait for the textures to be processed to continue with the game (input, audio, network, etc.)

Log in or register to post comments