GIMPS on PS3 - Page 2

Uncwilly · 2009-01-13, 01:53

Quote:

Originally Posted by flouran

Then, do you know how to patch MPrime for Cell?

You don't. You use MLucas or GLucas, they can be compiled for various processors. mprime/Prime95 have x86 family specific code.

flouran · 2009-01-13, 02:09

Quote:

Originally Posted by Uncwilly

You don't. You use MLucas or GLucas, they can be compiled for various processors. mprime/Prime95 have x86 family specific code.

Then basically I would have to compile Mlucas under YellowDog (installed on PS3) in order for it to work?

Also, does anyone know how I can run Prime95 on a Windows XP AMD Athlon x64 processor (as I have found an excellent computer)?

flouran · 2009-01-13, 02:38

I bet if you installed an x86 emulator for the Cell then we can run MPrime. I just don't know of any x86 emulators for the Cell processor architecture... Any ideas?

Uncwilly · 2009-01-13, 03:07

Quote:

Originally Posted by flouran

Then basically I would have to compile Mlucas under YellowDog (installed on PS3) in order for it to work?

Yes, you would have to make sure all of the libraries were suited to that install.

Quote:

Also, does anyone know how I can run Prime95 on a Windows XP AMD Athlon x64 processor (as I have found an excellent computer)?

AMD chips (Athlon, Operton, K6, etc.) are all in the x86 'family' and Prime95/mprime is ready to go on those. I ran Prime95 on an Athlon for several years.

flouran · 2009-01-13, 03:22

Quote:

Originally Posted by Uncwilly

Yes, you would have to make sure all of the libraries were suited to that install.

AMD chips (Athlon, Operton, K6, etc.) are all in the x86 'family' and Prime95/mprime is ready to go on those. I ran Prime95 on an Athlon for several years.

So basically, this weekend (or the next) I will attempt to compile MLucas. Note the word "attempt". If you guys want to help me out you are welcome to b/c I mean the primary success of Folding@Home was due to its port to the PS3 systems, so if I/we can install on Yellowdog, then we have made a landmark in GIMPS history.

CRGreathouse · 2009-01-13, 03:28

Good analogy, sdbardwick.

Quote:

Originally Posted by flouran

I mean the primary success of Folding@Home was due to its port to the PS3 systems,

Folding@Home doesn't do bignum operations: the Cell is perfectly suited for it. Glucas/Mlucas/Prime95 do major bignum operations with L-L, so the Cell is poorly suited for that. But regardless, I await the results (if you're able to compile).

flouran · 2009-01-13, 03:35

Quote:

Originally Posted by CRGreathouse

Good analogy, sdbardwick.

Folding@Home doesn't do bignum operations: the Cell is perfectly suited for it. Glucas/Mlucas/Prime95 do major bignum operations with L-L, so the Cell is poorly suited for that. But regardless, I await the results (if you're able to compile).

I'll try my best (and perhaps as a last resort will use an x86 emulator for Cell (if there is one)). Just a question: There is an option for merging CPUs, what are the possible risks and/or advantages associated with enabling this option? And should I do it if I have two or more computers (can somebody tell me from experience, etc.)? Also, if I select the option: "L-L test for world record" does it test the largest Mersenne prime already found or does it attempt to find another undiscovered world record prime?

What I am truly asking is what is the best option to choose in order to increase the probability of discovering a previously unknown Mersenne prime (I have two computers that are running Prime95 and MPrime currently)?

CRGreathouse · 2009-01-13, 04:19

For Prime95 (on your Athlon):
If you want to maximize your chances for finding a new Mersenne prime, do only L-L work (no factoring or ECM), and run one test per core (not across several cores).

Obviously the advice has nothing to do with your Cell experiment.

starrynte · 2009-01-13, 04:20

Also though the Cell may not be able to run LL/DC tests very well, I am interested how it does TF.

flouran · 2009-01-13, 04:26

Quote:

Originally Posted by CRGreathouse

For Prime95 (on your Athlon):
If you want to maximize your chances for finding a new Mersenne prime, do only L-L work (no factoring or ECM), and run one test per core (not across several cores).

Obviously the advice has nothing to do with your Cell experiment.

So should I set the master work type to LL-NF (LL without Factoring), LL (LL first test), D (LL double test), LL-WR (LL test for world record), LL-10M (LL test for 10+ million digits), or LL-100M (LL test for 100+ million digits)? Also, should I join a team or not?

CRGreathouse · 2009-01-13, 04:45

LL-NF. If you want to maximize your chance to find a Mersenne prime, you want to search the smallest numbers possible -- larger numbers are both harder to test and less likely to be prime. Further, if you can have someone else do the TF work, that makes your throughput on LL higher.

Joining a team or not doesn't affect your chance to find a Mersenne prime.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
128-bit OS'es and GIMPS?	ixfd64	Software	22	2011-10-31 22:23
GIMPS Nub	SayMoi	Information & Answers	5	2009-04-06 15:29
GIMPS uses only 1 cpu	Unregistered	Information & Answers	7	2009-01-10 20:01
GIMPS should pay	Vijay	Lounge	40	2005-07-01 18:10
Why do you run GIMPS ?	Prime Monster	Lounge	12	2003-11-25 19:04

2009-01-13, 02:38	#14
flouran Dec 2008 833₁₀ Posts	I bet if you installed an x86 emulator for the Cell then we can run MPrime. I just don't know of any x86 emulators for the Cell processor architecture... Any ideas?

2009-01-13, 04:19	#19
CRGreathouse Aug 2006 1011101011011₂ Posts	For Prime95 (on your Athlon): If you want to maximize your chances for finding a new Mersenne prime, do only L-L work (no factoring or ECM), and run one test per core (not across several cores). Obviously the advice has nothing to do with your Cell experiment.

2009-01-13, 04:20	#20
starrynte Oct 2008 California 2²×59 Posts	Also though the Cell may not be able to run LL/DC tests very well, I am interested how it does TF.

2009-01-13, 04:45	#22
CRGreathouse Aug 2006 3·1,993 Posts	LL-NF. If you want to maximize your chance to find a Mersenne prime, you want to search the smallest numbers possible -- larger numbers are both harder to test and less likely to be prime. Further, if you can have someone else do the TF work, that makes your throughput on LL higher. Joining a team or not doesn't affect your chance to find a Mersenne prime.