Fastest US Supercomputer Runs Linux
jgercken writes "The Department of Energy's Pacific Northwest National Laboratory has brought online a 11.8 teraflops supercomputer based on the Linux operating system, comprised of ~2,000 Itanium processors, and assembled by HP. Touted to be the fastest unclassified computer in the US, its main duties will be atmospheric chemistry, systems biology, catalysis and materials science."
If you notice this is based on PEAK Performance, aka Theoretical Max, not the best they've gotten out of it . . .
Your hair look like poop, Bob! - Wanker.
Also after reading the whole press release I'm stuck with a few measly pictures of a bunch of HP rack servers running a processor that I won't be able to buy (let alone afford) for awhile longer. There is no mention about how much heat the thing produces, or how much energy it takes to run it. I hope the Ph.D.s running the whole thing realize that while they are trying to do stuff for the "Department of Energy" they are releasing so many thousands of pounds of junk in the land/air/water to run this giant supercomputer.
Makes you wonder for a second what they're doing with the [SECRET - EYES ONLY] hardware.
My guess is that they're working on NP-hard, but useful problems, like finding ways to crack hard encryption via shortcuts that work half the time.
The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
sorry to hear that HP had something to do with it.
I used to really like HP, they used to make great laserjets, great ink/paintjets, excellent scanners, etc...
But then they merged with Compaq and kicked out the founder. And now, the most evil of evils, they are sleeping with SCO...
Once again, sad to hear that HP was in on this...
The Itanium 2 draws about 124 watts.
The Opteron draws about 85 watts.
My math skills may not be that great, but it looks like the AMD chip uses 31% less power than the Itanium 2.
We've got a related project called 'GridShell' that may be of some interest to the readers. Basically, it gives a slick WebUI as a front-end to an AI Grid Computing interface. As an added bonus, we've included our new implementation of the AWESOME new programming language 'SequenceL', which will AUTOMATICALLY and INTELLIGENTLY PARALLELIZE and DISTRIBUTE ITSELF across pretty much any Grid of Grids or Clusters or SuperComputers or whatever. You can check it out now at http://gridshell.sourceforge.net
/. editors want to post a story about this?
Thanks!
-Will MacBraswald, Jr.
Creator, GridShell
PS - Why didn't the
Creator of RPerl, Scouter, Juggler, Mormon, Perl Monger, Serial Entrepreneur, Aspiring Astrophysicist, Community Organiz
The 89W figure for Opteron is the maximum for the 130nm node - i.e., probably at 2.4 GHz.
At current speeds the Opteron is consuming under 70W.
The P4 and the Itanium 2 both draw more power than AMD processors though. The AMD is hot joke went south when Intel dropped the cooler PIII processors.
machines paid for with public funds do not count as anything of note... ever.
too much politics and not enough "bottom line"
instead i prefer to study all the machines listed in www.top500.org
and then i discount all the sites running clusters paid for using taxpayer dollars
www.top500.org is a fascinating list.
Hehe, I guess that's what seti@home's *really* doing.
I'm joking, I think.
"Proudly Posting Without Reading The Article"
In the dim times there was one company called Cray that built big, expensive custom designed vector supercomputers. It took forever to build one so Cray could always insure they were profitable because they always new how many orders they had well in advance and could staff and spend appropriately and they were pretty much the only game in town.
One day SGI got tired of doing just fluffy graphics and built the MIPS R8000 which was probably the first really successful CMOS supercomputer on a chip. They completely carved up Cray from the low end up and eventually pushed them into a merger from hell that nearly destroyed both companies.
Around this time the Department of Energy had to give up setting off nuclear bombs to see if they actually worked and got in the business of funding these massive supercomputers mostly to simulate bombs and then some other stuff too. Unfortunately the DOE changed companies and architecture with each new contract. They managed to suck SGI, Intel, IBM, Cray, HP and countless others in to this prestige contest and I doubt its been particularly good for any of them. You see these are one off systems, that require a massive very custom engineering effort and the R&D effort seldom pays off. Its just not a good way to do business spending massive engineering effort when your usually lucky to sell one system. If you get a second one you usually have to start from scratch and do it all over again.
They are great for prestige and maybe some of the R&D effort does translate into the companies product line but, IMHO, I think a smart, well managed computing company wouldn't touch these with a ten foot pole. Microsoft sure doesn't seem interested in pouring any effort in to trying to land one of these contracts.
If the U.S. government had a clue they would find a way back to pouring all their money in to Cray to develop the specialized vector processors and find a new little Cray Jr. company to specialize in building the giant Linux clusters and encourage companies like IBM and HP to get out of this massive distraction from their core business.
it says here (http://www.theregister.co.uk/content/61/32523.htm l - couldn't see anything in the posted article on this) that they achieved this speed by upgrading from 1ghz to 1.5ghz itanium's. What did they do with all those old CPUs? and who paid for the upgrade PNNL or HP? Those things aren't cheap, they cost $800+ each.
I wonder how much more powerful it could of been if they designed it with Opteron processors or had waited until next summer when the release of XDDR Ram that runs in the 3ghz range would of been out.
Keep in mind as main memory speeds catch up with processor speeds and can easily run in 128 & 256 bit configurations that the signifigance of chip cache will become less and less. If the memory standards commttee's can keep memory speeds in line wiht processors then we can see some great advances in supercomputing. Along with cheaper processors due to the lack of onboard cache's since the processors would be able to use the main memory for such purposes.
Only stopgap into truely fast computing is the hard drive and that is quickly coming into it's solid state future as well.
I would guess at 2006 for 10ghz PC's with the only moving parts left being the dvd player and cooling systems which at that time will probably have to be more advanced than even liquid unless we make thsoe processors run at that speed with todays power outputs.