PDA

View Full Version : Supercomputers...




ph8te
Feb 20, 2004, 02:48 AM
I caught a newssnippet this morning as I was waking up, and maybe there is a real tech expert out there who can give me an answer to my question...

According to the news, the "Computing-Centre" in Germany has just purchased an NEC Supercomputer for 36 Million Euro. I think its the same one as they have in Japan with 13 Teraflops?!?. Anyway, I know that the Mac-Cluster was less than 10% of this price, with almost the same computing power.

What I want to know, is if the Mac-Cluster has a disadvantage compared to the standard supercomputers, and If not, why do these Supercomputers still get chosen over a Cluster?



janey
Feb 20, 2004, 03:42 AM
same computing power? not quite.
check this out: http://www.top500.org/list/2003/11/

yes, the supercomputer cluster at virginia tech was one of the cheapest, but it doesnt really exist anymore (it was dismantled and will be replaced with G5 Xserves).

ph8te
Feb 20, 2004, 03:47 AM
I know that the cluster at virginia-tech is being redone, and maybe not the same computing power, but I am mainly talking about the price, the Apple cluster was so - for want of a better word - "cheap", what was the catch? Why would other institutions in need of a supercomputer not follow the same route as the Apple-Cluster, what are the advantages - if any - of a supercomputer compared to a cluster?

takao
Feb 20, 2004, 09:58 AM
actually those "top 500" are only based on 2 numbers

there are always different computers for different tasks. period.
in some scenarios a very big cluster like the apple cluster is actually _slower_ than normal supercomputers like those big towers from cray/nec/hitachi/ibm...
the bandwith between CPUs/ I/O speed etc. sums up very fast...

in real life aplication for whom they are build they are pretty fast
example:
#64 Hitachi SR-8000-F1 from the Leibniz Rechenzentrum, Munich can easily beat computers which are in front of himbecause the hitachi has advantages over clusters

those 'supercomputer-benchmarks' are artificial made up numbers..in real life it depends on the application/algorithms which are used..

ph8te
Feb 20, 2004, 10:02 AM
so in essence, it depends on what you want to do with the computer... and if cash is low, and you have a large area, and a specific application which works well with clusters, then you go for the cluster?

janey
Feb 20, 2004, 01:41 PM
Originally posted by ph8te
so in essence, it depends on what you want to do with the computer... and if cash is low, and you have a large area, and a specific application which works well with clusters, then you go for the cluster?
sure why not.

thats why so many scientists and stuff are getting attracted to Apple. Cheap supercomputers :)

Makosuke
Feb 20, 2004, 05:22 PM
Originally posted by ph8te
so in essence, it depends on what you want to do with the computer... and if cash is low, and you have a large area, and a specific application which works well with clusters, then you go for the cluster? Exactly.

Some work works spectacularly well with clusters--meaning high-speed communication between machines in the cluster isn't vitally important, such as with the SETI@Home and Folding projects, which are easily broken into chunks. For these projects, you'd be a fool not to go with a much cheaper cluster, and for reference that's also roughly what the Top 500 benchmark measures--the raw sum of computing power. Other projects just don't break up as well, in which case a NEC Earth Simulator-sytle system is preferable (the only choice, really), despite its higher cost per raw GigaFLOP.

And I think the Virgina Tech cluster is still running, they're just working through replacing the G5s with XServes--if I understand the idea of the way they set it up correctly, then they can replace chunks of it and still have it run just fine. They might even be able to replace machines without shutting it down.

takao
Feb 20, 2004, 07:15 PM
if somebody is interest about the new supercomputer i just found some pictures

i hope it's the same as mentioned in the first posting

http://www.8ung.at/computernerd/cluster1.html
http://www.8ung.at/computernerd/cluster2.html
.
.
.

(link is in german ..but very nice pictures..)

price was 55 Million € and it was made for the "John Von Neuman Institut for Computing" (NIC) in Juelich

Makosuke
Feb 20, 2004, 10:12 PM
Sorry, Takao, it looks like the pictures you posted aren't from the same computer--it says that one is made up of 31 IBM p690 servers running Power 4+ processors, with a total max speed of about 9 TFLOPS, putting it at 12-14 on the world rank, according to that page. Cost of 55 Mil Euros, making it a whole lot more than the VT cluster cost, though this one probably handles non-parallell tasks somewhat better. Also features 50TB of storage online... woo.

If the original poster was right, then the new supercomputer should use NEC vector processors, and be a whole lot faster than that.

geerlingguy
Feb 21, 2004, 02:32 PM
I just wish I could have one of those Xserves to render me my FCX and DVDSP stuff.

takao
Feb 21, 2004, 04:00 PM
yeah but _that_ supercomputer was introduced the day the first posting mentioned new supercomputer in germany ...and if their were any new supercomputers bought by german universities etc. there are always news in german computer news tickers too (www.heise.de as a example)

perhaps it was mistaken because the supercomputer was _bought_ by the 'NIC' Institute (as mentioned before) and not made by 'NEC'

there aren't many supercomputers sold here ;)

the german 'Wetterdienst' (weather calculations etc.) got a NEC last June with additions later the year that was only news which i found after a quick google search

ph8te
Feb 23, 2004, 02:40 AM
Ok, since I was the original poster of that message, I have to make a correction, and blame the inaccurate journalists, who said that the machine had 13 Teraflops, cost in the region of 34 Million Euro, and was ordered from Japan.

After a bit of searching, I found the following link which refers to the story at hand, and also corresponds with the date when I heard the story...

http://www.fz-juelich.de/portal/home

... leave it to the journalists of this world to get things wrong again...