How to have all 4 cores working on a single number?

tech96 · 2014-07-03, 02:49

I recently installed Prime95 on my pc,and found out that it always assign four different numbers for the four cores.Is it possible to work on a single number using all four cores, if so how?

LaurV · 2014-07-03, 04:03

Go to test/workers, chose "1" single worker (the topmost box) and chose "x" for the number of CPUs (cores) (bottom most box) (where x is your number of physical cores, not the hyperthreaded cores, for example x=4 if you have 4 physical cores, like a i7-2600, no matter that you can use 8 HT cores, the HT does not bring any benefit for LL testing, except generating more heat and making your computer less responsive in some cases).

Restart P95 from Test/exit, then run again.

One more thing to do (optional) before restarting, would be to edit the worktodo file and move all the work into "worker 1" section, then delete the "worker x" lines for the other (now empty) sections. If you don't do this, you will get a warning telling you that "worktodo" contains more workers than expected, and there will be no work done for those sections. Caution: do not delete work! Just delete the [worker x] lines. This step is optional. If you are unclear, then do nothing. No harm done if you forget it, just some work which may be recycled in the future.

TheMawn · 2014-07-03, 04:33

Just be aware that using multiple cores for a single number is slower in the end. You're most efficient when using one core per worker.

LaurV · 2014-07-03, 05:01

True. Using multiple cores for the same assignment (multithread) is only justified when you want to finish some special assignment very fast (like impatience, or double-checking a mersenne which was found prime by the first LL test

) or if you have problems with the heat, memory bandwidth, etc. (like your computer can not sustain the memory transfer required by 4 workers, new cpus/mobos/chipsets should *not* have this issue!).

To give an example, if testing an exponent in some particular range, using a single core, takes 40 days, then testing it in 4 cores will take 11 days (and not 10 days!). In long run you will need 44 days to test 4 exponents one-by-one. It may temper your patience having an output every 11 days, but at the end is a gambler fallacy, you lost 4 days. You can achieve the "one output every 10 days" by using 4 workers, each one in one core, and starting with different sizes of exponents, such as when first worker is starting, the second one is 25%, the third is 50%, the last is 75%. You also can use "multithread" to achieve this "shifting", or start with exponents of different sizes, and after a while you achieve your "one LL output every 10 days" and calm your pride

Disclaimer: the numbers are only examples. You mileage may vary with the system you have and range you are testing.

Uncwilly · 2014-07-03, 05:42

Quote:

Originally Posted by LaurV

You can achieve the "one output every 10 days" by using 4 workers, each one in one core,

Or run 3 LL tests on 3 cores and have one core do P-1 testing. You will get a lot more outputs per day with P-1.

tech96 · 2014-07-04, 09:53

Thanks,it worked. And yeah its more like 3 times faster than 4 times on my pc.

2014-07-03, 02:49	#1
tech96 Jul 2014 2₁₆ Posts	How to have all 4 cores working on a single number? I recently installed Prime95 on my pc,and found out that it always assign four different numbers for the four cores.Is it possible to work on a single number using all four cores, if so how?

2014-07-03, 04:03	#2
LaurV Romulan Interpreter Jun 2011 Thailand 22613₈ Posts	Go to test/workers, chose "1" single worker (the topmost box) and chose "x" for the number of CPUs (cores) (bottom most box) (where x is your number of physical cores, not the hyperthreaded cores, for example x=4 if you have 4 physical cores, like a i7-2600, no matter that you can use 8 HT cores, the HT does not bring any benefit for LL testing, except generating more heat and making your computer less responsive in some cases). Restart P95 from Test/exit, then run again. One more thing to do (optional) before restarting, would be to edit the worktodo file and move all the work into "worker 1" section, then delete the "worker x" lines for the other (now empty) sections. If you don't do this, you will get a warning telling you that "worktodo" contains more workers than expected, and there will be no work done for those sections. Caution: do not delete work! Just delete the [worker x] lines. This step is optional. If you are unclear, then do nothing. No harm done if you forget it, just some work which may be recycled in the future. Last fiddled with by LaurV on 2014-07-03 at 04:06

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
does half-precision have any use for GIMPS?	ixfd64	GPU Computing	9	2017-08-05 22:12
Single exponent testing on MultiCore	graizada	Software	3	2013-02-05 14:36
Single LL Question	Unregistered	Information & Answers	18	2012-04-12 19:18
Large number of exponents assigned to single computer	dave_0273	Lounge	6	2004-03-28 07:19
4 checkins in a single calendar month from a single computer	Gary Edstrom	Lounge	7	2003-01-13 22:35

2014-07-03, 04:33	#3
TheMawn May 2013 East. Always East. 11·157 Posts	Just be aware that using multiple cores for a single number is slower in the end. You're most efficient when using one core per worker.

2014-07-03, 05:01	#4
LaurV Romulan Interpreter Jun 2011 Thailand 10010110001011₂ Posts	True. Using multiple cores for the same assignment (multithread) is only justified when you want to finish some special assignment very fast (like impatience, or double-checking a mersenne which was found prime by the first LL test ) or if you have problems with the heat, memory bandwidth, etc. (like your computer can not sustain the memory transfer required by 4 workers, new cpus/mobos/chipsets should *not* have this issue!). To give an example, if testing an exponent in some particular range, using a single core, takes 40 days, then testing it in 4 cores will take 11 days (and not 10 days!). In long run you will need 44 days to test 4 exponents one-by-one. It may temper your patience having an output every 11 days, but at the end is a gambler fallacy, you lost 4 days. You can achieve the "one output every 10 days" by using 4 workers, each one in one core, and starting with different sizes of exponents, such as when first worker is starting, the second one is 25%, the third is 50%, the last is 75%. You also can use "multithread" to achieve this "shifting", or start with exponents of different sizes, and after a while you achieve your "one LL output every 10 days" and calm your pride Disclaimer: the numbers are only examples. You mileage may vary with the system you have and range you are testing.

2014-07-04, 09:53	#6
tech96 Jul 2014 2 Posts	Thanks,it worked. And yeah its more like 3 times faster than 4 times on my pc.