|
Title: Yet Another SuperComputing Mac Cluster Post by: BFG on June 22, 2004, 11:24:00 am Quote COLSA, a U.S. Army contractor based in Huntsville, Alabama, today announced the purchase of 1,566 Xserve G5s to build a new supercomputer expected to be one of the fastest in the world. The supercomputer, named "MACH 5," is expected to deliver a peak performance capability of more than 25 TFlops/second at a cost of US$5.8 million and will be used to model the complex aero-thermodynamics of hypersonic flight for the US Army. How cool is that! 25TFlops should stick it in the top three with ease as far as i can remember, although IBM are currently working on a supercomputer thats probably going to knock the earth simulator off the top spot in a couple of years! Title: Re:Yet Another SuperComputing Mac Cluster Post by: Toxic::Joka on June 22, 2004, 11:43:08 am IMB ???
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 22, 2004, 12:12:54 pm meh, one little typo and your up against the wall and the firing squad are taking aim ;)
Title: Re:Yet Another SuperComputing Mac Cluster Post by: Civrock on June 22, 2004, 12:30:25 pm IBM... hmm... does that mean it?s going to be with G5s? a supercomputer by IBM themselves?
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 22, 2004, 01:49:03 pm Quote The latest edition of the Top 500 list of the world's fastest supercomputers reveals that Japan's Earth Simulator supercomputer, built by NEC, retains the No. 1 spot for the third year in a row, while a new system called "Thunder" claims the No. 2 position. It is an Intel Itanium2-based cluster system at Lawrence Livermore National Laboratory (LLNL) in California. In addition, No. 4 and No. 8, prototypes for the upcoming IBM BlueGene/L system, make the list for the first time. The prototypes are a joint development of and IBM and are currently at IBM's facility. The final system will be installed at LLNL and is expected to replace the Earth Simulator by June 2005 or by the end of 2004. Dosn't sound like G5's... But rather somthing custom betweeen LLNL and IBM. - seems daft though considering the power to cost ratio of the G5 Xserves - Hell look at the difference of the cost of this new Xserve one at 5.8Mill conpared to the 350Mill Earth Simulator - yet the power difference is so small! Title: Re:Yet Another SuperComputing Mac Cluster Post by: Mr. Lothario on June 22, 2004, 01:52:24 pm Yeah, but if it is something custom, that's R&D that IBM is being paid for. That's good for the industry.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 22, 2004, 07:00:14 pm True, very true, and you never know it might just trickle its way down into the G6 ;)
Amazing to think how cheap you can get a seriouse supercomputer now though compared with say, 5 years ago Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 22, 2004, 07:08:14 pm http://www.macrumors.com/pages/2004/06/20040621202310.shtml
That sums it up a lot better. I think we now know where the G5 Xserve supply has been disappearing to. Title: Re:Yet Another SuperComputing Mac Cluster Post by: Supernatural Pie on June 22, 2004, 08:33:42 pm Hypothetically -
How fast could a supercomputer made up of X-Serve G5's be if they spent the same amount of money as the really expensive clusters? (?350 mil) Title: Re:Yet Another SuperComputing Mac Cluster Post by: crypt on June 22, 2004, 08:36:00 pm i dunno snipe, maybe fast enough to run rvs, maybe more. probably less though
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 22, 2004, 09:16:25 pm Foolish, foolish Crypt. Snipe, ask Steve Jobs at WWDC.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 22, 2004, 09:36:56 pm meh we should be able to work that out. If we divide the cost of the Virgina tech computer by the number of nodes, then multiply by the closest denominator to reach 350million, then take that number and.... oh shit im lost. but its definatly easy to do... isn't it?!
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 22, 2004, 09:51:25 pm Objection on the grounds BFG doesn't know what he's talking about.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 23, 2004, 12:24:22 am Sustained. I hate maths, just give me 5 mins with a base ball bat and long division....
oh btw. some more info on the cluster here (http://www.macminute.com/2004/06/22/supercomputer) damn it im gonna get a calculator and work it out now :D Title: Re:Yet Another SuperComputing Mac Cluster Post by: c| Lone-Wolf on June 23, 2004, 06:13:12 am I ran the numbers on the Virginia Tech supercomputer, way back a whole long time ago.
Baring everything that i dont know about (which is everything) it was some absurdly more powerful computer, for the same 350 million cost. It really isnt that hard, i mean, how many Tflops can the VT computer pull now? Divide 350 million by the cost of the Virginia Tech supercomputer, then multiply that number by the number of TFlops the VT computer can do right now. And you have your answer. 350/5.5(?) = 63.5 (?) 63.5 x VT SuperComputer Terraflops = ? Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 23, 2004, 06:15:26 am Well, are you gonna leave us with that cliff hanger lone or do the math?
Title: Re:Yet Another SuperComputing Mac Cluster Post by: Ace on June 23, 2004, 08:28:25 am His math is irrelevant because these clusters will not scale linearly.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 23, 2004, 11:13:58 am Sure Ace, but it gives a ball park figure, be it a very big ball park ;)
Title: Re:Yet Another SuperComputing Mac Cluster Post by: c| Lone-Wolf on June 23, 2004, 01:20:36 pm Baring everything that i dont know about (which is everything) Ace, i was running the numbers ignoring the fact that they wouldnt "scale linearly" as you say. I knew something like that would probably make my calculations moot. But Ace, what would happen, would the end computer for 350 million be more powerful than what the numbers were that i came up with? Or would it be slower? And by your guestimation, how much in each way? Cmon mr. smart guy, enlighten us laymen. ;) (Btw, i love your icon Crypt) Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 23, 2004, 03:19:42 pm guestimation.... that sounds like a Dubbya word.
" ba ma guestimation i have been thinkingamafying that our treatmentation of prisoners is justifickable ;) Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 23, 2004, 04:19:44 pm Well, Xserves are cheaper than Xeons and the other fancy computers, and they put out as much, if not more, power.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: Cobra on June 23, 2004, 08:32:15 pm But Ace, what would happen, would the end computer for 350 million be more powerful than what the numbers were that i came up with? No. That would be scaling exponentially. Which only occurs in Koala-Walla Land. They would scale less than linearly, so your calculations would be too high.Title: Re:Yet Another SuperComputing Mac Cluster Post by: BTs_Mysterio on June 23, 2004, 09:08:41 pm But Ace, what would happen, would the end computer for 350 million be more powerful than what the numbers were that i came up with? No. That would be scaling exponentially. Which only occurs in Koala-Walla Land. They would scale less than linearly, so your calculations would be too high.1x1 is 2...I like math. Title: Re:Yet Another SuperComputing Mac Cluster Post by: BFG on June 23, 2004, 09:16:20 pm it will be less than a 1x1 direct scale - as power must be lost as the cluster grows - it can't be a pure scale upwards without some loss... i guess it all depends what the scale of that loss is as the cluster grows.
Title: Re:Yet Another SuperComputing Mac Cluster Post by: Ace on June 23, 2004, 11:08:18 pm Baring everything that i dont know about (which is everything) Ace, i was running the numbers ignoring the fact that they wouldnt "scale linearly" as you say. I knew something like that would probably make my calculations moot. But Ace, what would happen, would the end computer for 350 million be more powerful than what the numbers were that i came up with? Or would it be slower? And by your guestimation, how much in each way? Cmon mr. smart guy, enlighten us laymen. ;) (Btw, i love your icon Crypt) It would definitely come out with less power than the numbers you were coming up with. Basically, the more nodes you add to a cluster, the lower your TeraFlop/node average will be for the cluster as a whole. To be honest, I couldn't even come close to providing a reasonable guess as to what kinda performance you would get out of $350 million worth of XServe G5s. It would depend on what interconnects they use and how they lay out the cluster network. The only thing I can say for sure is that it would eat the Earth Simulator alive. |