Single computer breaks PetaFLOP barrier

[Michael_McCord,_M.D.]

From today's Houston Chronicle-a US military supercomputer codenamed Roadrunner, became the first single computer to break the PetaFLOP barrier at 1.026 quadrillion calculations per second. It is located at Los Alamos, NM. How much calculating power is that? If all the 6 billion people of the world used hand calculators 24 hours a day, it would take us all 46 years to do what this computer can accomplish in 1 day.

FAH has the combined power of TWO of these!! and growing.......

[VijayPande]

I know the specs of this machine well. It's similar to FAH in the sense that it has a lot of Cells and regular CPUs, although less than FAH.

[Evil Penguin]

I'm sure it's a fairly expensive computer.
Not to worry, new faster hardware comes out every 6 months or so.

[Michael_McCord,_M.D.]

116,600 cores would require a heckuva SMP client! haha and 3 Megawatt power drain-you would have to live next door to the Hoover Dam-or a big nuke plant!

[VijayPande]

116,600 cores would require a heckuva SMP client! haha and 3 Megawatt power drain-you would have to live next door to the Hoover Dam-or a big nuke plant!

For Roadrunner, we'd run SMP cores to max scaling, which would likely be ~32 to 256 cores for the A2 binary. We'd then run multiple jobs to fill out the rest.

[zanth]

Wouldn't that be incredibly cool to get to try it out?

[sneakers55]

it's a struggle to save up anything for the next upgrade to a PC, rather than having to save up enough for one tank of gasoline. Perhaps the RoadRunner will study alternative forms of energy, who knows?

Department of Energy has it, it'll be used to study nukes.

Enormous amounts of calculation to answer a question (will the aging nukes still work?) that could be settled more easily by setting off a few bombs safely underground. Before you wonder too much DOE spends around $4 billion a year to answer this question.

I just wish more people would join the F@H project.

Don't we all?

[bruce]

. . . with the whole country going broke, we'll just have to confine ourselves to speculation and dreaming. . . .

OK since you've already sidetracked the thread with politics, maybe the government (or Los Alamos) would be willing to donate any free time they have to FAH.

, , , no, not really. We don't want FAH to be supported by the military side of the government so that the results can be placed in the public domain by Stanford rather than by the government.

116,600 cores would require a heckuva SMP client! haha and 3 Megawatt power drain-you would have to live next door to the Hoover Dam-or a big nuke plant!

I'm sure Los Alamos has it's own nuclear power generation capability.

[toaster8]

Roadrunner is world's fastest computer

Updated Mon. Jun. 9 2008 2:44 PM ET

The Associated Press

WASHINGTON -- Scientists at the Los Alamos government weapons lab have built the world's fastest computer. It is capable of sustaining 1,000 trillion operations per second.

The Energy Department an IBM Corp., announced the breakthrough Monday. The computer will be used to help maintain the nation's nuclear weapons stockpile. Energy Secretary Samuel Bodman says the new computer, named Roadrunner, also will be used to help solve global energy problems and "open new windows of knowledge" in basic research.

The Los Alamos National Laboratory in New Mexico and IBM worked for six years to achieve the world record computer speed.

[Zch]

If donated, what're the chances F@H could get a day or so? If so, are there any problems that RoadRunner could give us a shortcut on? Darn, what Stanford/F@H could do with RoadRunner in just a years time, plus all their distributed computing assets! What the whole world needs is more computing power, each and every day.

As the practical benefit of computer simulations becomes more evident in growing fields of research and especially economically advantageous in some of them, more and more computers will be coming.

I also wonder what could Folding@home do with RoadRunner. Would a day of computation on its 1 PFLOPS deliver a supplement of merely half a day to ~2 PFLOPS powered Folding@home or rather provide some brand new capacities due to architectural differences.

[alancabler]

Some of you know that 1 supercomputer tried a brief test of FAH in the past year (Fermilab).
The performance was astounding, to say the least.

note: No official statement was made by either Pande Group or F.N.A.L.
The test was deduced by FAH donors.
Such is the life of a mushroom farmer.

[alancabler]

Greetings Adam,

I wish I knew the results of that test, if it took place.

(Linked in my post...) http://folding.extremeoverclocking.com/ ... s=&t=48936
here's another link: http://fah-web.stanford.edu/cgi-bin/main.py

That result (4 Million+ pts) was from just a few days... about 4 days I think, forgot the exact time frame...
When the scores were "fresh" and the stats pages gave more detail, you could see it ramp up and back down
like a bell curve, or a square-wave.

No one who knows anything has said anything- Dang!

[dittopb]

The result is probably not worth a Nobel prize.

[alancabler]

Sheer speculation:
Fermilab output points-wise was about the same as team Overclockers Australia, with about 1/4 of Team OA's daily WU production. Since Team OA contains a mix of standard CPU clients, SMP, PS3 and GPU clients, their pts/WU ratio would be lower than team Fermilab- If Fermilab's effort was confined to high-point-value SMP WUs.

One could guess that Fermilab was involved with Pande Group in an experiment regarding SMP scaling, probably also in concert with Stanford's Pervasive Parallelism Laboratory.

Wild-@$$ guess #2:
Fermilab's effort was nothing officialy coordinated, they just had to shut down their accelerator for a few days to sweep out the cigarette butts, or whatever, so their mainframe wasn't busy and they decided (down at the pub) to take 'er out for a spin and see what she'd do...
Nah, that can't be it.
Can it?

[dittopb]

i like the wild@ss guess #2.