He isn't starting the exponent from the beginning. He moved the save file to the new computer.

[quote=Primeinator;177199]He isn't starting the exponent from the beginning. He moved the save file to the new computer.[/quote]

Ahhh ok ... still I need some of these Nehalem EX. I've read you can combine up to 8 of them on a mainboard. Using 64 real cores with prime95 would be fun.

[QUOTE]Ahhh ok ... still I need some of these Nehalem EX. I've read you can combine up to 8 of them on a mainboard. Using 64 real cores with prime95 would be fun.[/QUOTE]

No kidding... till you reach the point where it no longer scales. I would love to have one of those things but I don't think I would like paying the electricity bills.

[quote=Primeinator;177201]No kidding... till you reach the point where it no longer scales. I would love to have one of those things but I don't think I would like paying the electricity bills.[/quote]
8 processors with 130 Watt each ... isn't that expensive (perhaps you get an extra rate with your power plant ;). If it won't scale I'll send another one of these babies to George so he can optimize it for high core usage.

Unfortunately until I'm rich that have to wait. ;).

Just a word of warning... as Kevin's result has possibly shown, x86 CPUs are susceptible to silent errors.

Processors like Itanium, Sparc, Power have significant portions of silicon dedicated to error monitoring and correction. The SPARC64 VII CPUs I am using have just over 2000 different checks happening continuously.

Silent errors occur fairly often on x86 chips. In many cases a wrong pixel isn't noticed but we've all experienced BSOD which often can be attributed to silent errors. You will also be surprised at the high rate of silent errors on disk drives (very very scary).

Now for the good news... with Intel finally bringing the memory controller on chip and implementing QPI the resulting Nehalem processor has good Ghz, good IPC, has decent on chip cache, has a nice core count and scales well to several sockets .

Glucas & Mlucas use different threading models so there's some fun times ahead with making the most of these new processors using different compilers, operating systems and hardware combinations.

Certainly we need to make some advances. I did a basic calculation (making lots of spurious assumptions) and doing one run of LL for a 100 million digit Mersenne Prime today would take 135 years using 16 cores!!!

[QUOTE=rgiltrap;177219]I did a basic calculation (making lots of spurious assumptions) and doing one run of LL for a 100 million digit Mersenne Prime today would take 135 years using 16 cores!!![/QUOTE]Are you sure of that figure? It doesn't seem that you have a very good core if that is that case.

It seems the current Quads of today can do a 100Mdigit MP in just a few years. So your 16 cores must all be really crappy to take 135 years! :evil:

[QUOTE=ATH;177172]But I doubt its possible Sq = 0 (mod 2^p-1) with q<p-2.[/QUOTE]If Mp is composite, I think it is possible... In fact, we know nothing about the DiGraph under x^2-2 modulo a Mersenne composite... except that, with Mp composite and S0=4 then Sp-2 cannot be 0.
Tony

[QUOTE=rgiltrap;177219]Certainly we need to make some advances. I did a basic calculation (making lots of spurious assumptions) and doing one run of LL for a 100 million digit Mersenne Prime today would take 135 years using 16 cores!!![/QUOTE]

The number seems to be consistent with a 1 billion digit mersenne.

[QUOTE=rgiltrap;177219]... doing one run of LL for a 100 million digit Mersenne Prime today would take 135 years using 16 cores!!![/QUOTE]??? I found only 5 years.
(15 days for a fast PC) x (10 times more digits) x (12 times more computation per iteration) ~= 5 years.
How did you compute ?
Tony

[QUOTE=axn;177228]The number seems to be consistent with a 1 billion digit mersenne.[/QUOTE]

Mea culpa, mea culpa, mea maxima culpa. You are correct I meant 1 billion digits. :redface:

Date/Time of result of my verifications
 

Glucas/Nehalem : end today ~2:00pm USA CST (9:00pm France) ~90% done now.
Glucas/Itanium2 : end sunday ~0:20am USA CST (7:20pm France) ~80% done now.
I see with George how he wants the results to be announced.
("Jusqu'ici, tout va bien...")
Tony