petaflop computer built

[Fusion_power]

http://news.bbc.co.uk/2/hi/technology/7443557.stm

Wonder what one of these would do for gimps.

DarJones

[Mini-Geek]

As of latest count, GIMPS has about 27 TFLOPS. Assuming this is all double precision (it's based on the PS3's type of CPU so I'm not sure), this machine is over 37 times as powerful as GIMPS. If you could write a GIMPS client for it, you'd find primes 37 times faster than we do now. We've been finding them about every year, but we've been lucky. Taking a pessimistic view that we should average one every, say, four years (anyone know what we "should" average?), this would bump it up to about every 39 days. One year average GIMPS would be about 10 days per prime, average.
I wonder if it would be cost-effective for them to run Prime95 until they found a 10M digit prime (their cut of the prize money vs the electricity and man-hours to setup that many clients and run the supercomputer). It wouldn't take them long. Too bad we can't get them to run GIMPS on that in its down time...we could probably have a 100M digit prime within a few years, max...anyone want to hack into it and see if they notice?

[Uncwilly]

Just let them do all of the factoring to 5 or 6 extra bit levels.
Imagine the P-1 that it could do.

[ewmayer]

Originally Posted by Uncwilly

Just let them do all of the factoring to 5 or 6 extra bit levels.
Imagine the P-1 that it could do.

TF to 6 more bit levels would find only a few % more factors for ~64x more work. In terms of overall project throughput P-1 is barely a breakeven strategy even with carefully set stage bounds.

The most useful work such a machine could do is LL testing just like our humble peecees do - but the whole point of a project like GIMPS is to use cycles that would be otherwise idle to instead do some "useful" [or at least interesting] work. Folks don't spend multimillions on supercomputers to have them sit idle.

Of course, with a beast like this even a few % spare cycles is a lot compared to the legion of peecees, but I kinda like the "empowering the humble rabble" aspects of the project-as-is.

[retina]

The thing is for military use only anyway. So projects like GIMPS will be so far down the list of things to compute that I doubt it will make it into the top zillion. More likely something to work out the most efficient way to kill the enemy. Hehe, let's hope it has a good reciprocation algorithm, I can see War Games is on the way soon. Anyone want to play tic-tac-toe, or maybe global thermonuclear war instead?

[petrw1]

Originally Posted by retina

The thing is for military use only anyway. So projects like GIMPS will be so far down the list of things to compute that I doubt it will make it into the top zillion.

Exactly ... with all those zillion other things to worry about no one would ever find GIMPS running on it.

Anyone have a friend there?

[ixfd64]

A long time ago, I recommended sending e-mails to organizations that use such computers, but most members of the forum opposed were to the idea and said that it would be "spamming." :\

[jasonp]

Originally Posted by ixfd64

A long time ago, I recommended sending e-mails to organizations that use such computers, but most members of the forum opposed were to the idea and said that it would be "spamming." :\

It is spamming. What would you think of people who posted here wanting GIMPS users to switch to cancer/medical projects, who insisted that it was a good cause and that if you really cared about people with cancer then you would stop wasting your time with silly math projects? IIRC this happened a lot with folding@home...