Holy new Mersenne prime, Batman! (M47 related)

[Orgasmic Troll]

Originally Posted by R.D. Silverman

I approve of the stated goals of EFF. Now reduce my ignorance.
Tell me the scientific (or even mathematical) problem that is being solved
by actually finding Mersenne Primes.......

I'm sorry, but what does that have to do with anything? I have to guess at what you're getting at here, which I think is that if the EFF goal is to solve scientific problems, then why is it throwing money at finding Mersenne Primes if they don't solve any scientific problems?

If this is what you meant, then I have to point out that the money is not for finding a mersenne prime, but just a prime over 10m digits. Sure, mersenne primes were the sure bet for getting there eventually, but that doesn't preclude someone from finding a new way of proving primality that gives LL a run for it's money (semi-literally)

[ixfd64]

Originally Posted by retina

Still hasn't caught the 41st Known Mersenne Prime Reported!! thread. But it won't be long till it does.

The current leader is New to GIMPS? Start here!!!, 49467 views.

Somebody is cheating! 122803 views for this thread now!

How on Earth did this thread suddenly gain 100,000 views? I hope it wasn't because of someone using a bot to automatically grab pages. That would hog way too much resources.

I have a feeling that Mike simply prepended a "1" to the number.

[sylvester]

Originally Posted by Orgasmic Troll

Then why run GIMPS?

Since nobody seemed to touch the position of people like me, I need to give you an explanation.

GIMPS has stumbled upon a way of extracting a usefull work out of continuous testing of machine's motherboard. I don't really care about Mersenne primes. Up till 2003 when I joined GIMPS I used to simply repeatedly run a CPU&memory intensive job for which I knew the exact expected answers.

For me and many people like me GIMPS is just a side effect of an effort to do the quality assurance on my own service. Just this year prime95/mprime discovered two bad ASUS motherboards (bursted capacitors) that were in continuous operation for over 4 years. In 200[67] GIMPS discovered the same problem in a computer located in a very remote data center, where I would have no other sensible way to test or inspect the machine. The savings on shipping and labor I got on those 3 occasions more that outweight the additional cost of electricity I paid to run the GIMPS on 10-20 machines that are under my care.

For now you guys are looking forward to get-togethers and beer drinking. Enjoy! But there will be a day when you will again start looking into how to get more people to run the GIMPS client. Here are my top 2 suggestions:

1) SNMP trap support in addition to prime.log. I discovered that the machines broke when all of sudden the CPU load on them decreased below 100%. The preferred way to report the errors in the enterprise is SNMP. Supporting SNMP may not give you creed amongst the mathematicians, but it will give you instant creed amongst the IT managers.

2) Develop a workaround for a problem where SQL servers running on the same machine as GIMPS keep postponing the maintenance tasks that are scheduled for the periods of low CPU load. This affects MSSQL 2005 SP2 EE, maybe others.

I doubt I will see any action on this front, but at least I tried to state the business case.

[henryzz]

http://www.mersenneforum.org/showthr...ewpost&t=10313
this project was calculated to have a 1/26th chance of finding a 10million digit prime

[ewmayer]

Regarding parallel scalability [which Tony Reix also asked me about offline], here are the numbers for Mlucas on the 2.1 GHz Sparc VI [dual-core CPUs] M5000 server, at an FFT length of 4096K:

#threads    iteration    speedup vs    CPU
            time (sec)   1-thread      Utilization
--------    ---------    ----------    --------
 1          0.237         1.00x        100%
 2          0.138         1.72x         86%
 4          0.065         3.65x         91%
 8          0.038         6.24x         78%
16          0.022        10.8 x         68%

Interesting what happens at 2 and 4-threaded ... CPU utilization drops to 86% at 2-threads, then rises to 91% at 4 threads before dropping again.

Also, I should point out that on at least one architecture [Itanium2], Mlucas actually showed a slightly *superlinear* scaling for up to 4 threads [see here, especially post #19 in the thread], likely due to the working set size per thread for 2-4 threads fitting into each processor core's cache better than the full working set size for the FFT being done. So there is in fact potentially an absolute throughput gain to be had from running multithreaded. Not that such information is "useful" or "interesting" or anything...

[patrik]

Originally Posted by jinydu

On the other hand, it might have been nothing more than luck. How about you try reporting it a few more times?

If it still doesn't work after a few more attempts... What error message do you get?

The result was accepted after a few hours. The error was Error 3. Something like "Database crashed or exponent not assigned to you".

[Prime95]

Originally Posted by ixfd64

In Cooper and Boone's talks on M43 and M44, they mention that the top reason for finding Mersenne primes to win the EFF prize.

They may have said that to spur interest in GIMPS. If they really believed that then they would have been testing only 10M digit numbers all these years.

[ixfd64]

Originally Posted by Prime95

They may have said that to spur interest in GIMPS. If they really believed that then they would have been testing only 10M digit numbers all these years.

Good point. I guess we all have to make white lies from time to time.

[sylvester]

Originally Posted by Prime95

Hypothetically speaking, what would you propose for the $150K award that is probably 15 years away.

The way the currencies and central banks are doing nowadays would probably mean that in 15 years $150K will be just enough for a couple of postcard stamps and maybe a pitcher of beer.

[T.Rex]

Originally Posted by rgiltrap

... we have identified two new obscure bugs in SunStudio.

Compiling Glucas with the 2 last versions of icc on Itanium2 also has shown problems...
Tony

[R.D. Silverman]

Originally Posted by Mini-Geek

Why find Mersenne Primes?
As Kirk might put it (if he were interested in primes instead of mountains), "because they're there."

Now, that's a good reason!!!