Trial division with CUDA (mmff) -- used, but runs like new! - Page 13

ET_ · 2012-09-11, 09:12

Quote:

Originally Posted by ATH

Testing for best GPUSievePrimes for fermat numbers with GTX 460 on a Q9450 (Yorksfield) 2.66 Ghz:

Code:

			     Best
 	  		GPUSievePrimes	 Speed
FermatFactor=30,x,x	   70k-140k	883.51 M/s
FermatFactor=40,x,x	  125k-160k	760.80 M/s
FermatFactor=50,x,x	  220k-270k	385.49 M/s
FermatFactor=60,x,x	  300k-310k	316.46 M/s
FermatFactor=70,x,x	  280k-360k	289.77 M/s
FermatFactor=80,x,x	  400k-580k	173.17 M/s
FermatFactor=90,x,x	  450k-500k	148.47 M/s
FermatFactor=100,x,x	  400k-570k	141.48 M/s
FermatFactor=125,x,x	  630k-840k	78.01 M/s
FermatFactor=150,x,x	  650k-1075k	61.62 M/s

Waiting for similar tables on other GPU models, to optimize ranges.

Please,

Luigi

LaurV · 2012-09-11, 12:35

Quote:

Originally Posted by LaurV

When we reach home tonight we will post the result.txt, which is not accessible now

as promised:
(finished for MM107, I saw it is already marked on the list as finished, and partially done for the mm61, split in two semi-ranges, each semi-range split in - now 11 remaining - parts, each part would take about 9 hours and half on a gtx580 a bit overclocked and water cooled :D).

Code:

[Tue Sep 11 07:35:46 2012]
no factor for MM61 from 2^114 to 2^115 (k range: 5000T to 5100T) [mmff 0.21 mfaktc_barrett120_M61gs]
[Tue Sep 11 05:44:13 2012]
no factor for MM107 from 2^155 to 2^156 (k range: 0T to 200T) [mmff 0.21 mfaktc_barrett160_M107gs]
[Tue Sep 11 15:25:08 2012]
no factor for MM61 from 2^114 to 2^115 (k range: 6300T to 6400T) [mmff 0.21 mfaktc_barrett120_M61gs]
[Tue Sep 11 19:24:53 2012]
no factor for MM61 from 2^114 to 2^115 (k range: 5100T to 5200T) [mmff 0.21 mfaktc_barrett120_M61gs]

With flashjh's settings (816000,128,16) this scores between 611M/s and 618M/s and it seems to be close to the optimum for the card/expo/range combination. The computer can't be used (almost, but nobody need it right now).

flashjh · 2012-09-11, 13:31

Quote:

Originally Posted by LaurV

The computer can't be used (almost, but nobody need it right now).

(forgot to mention that part

)

axn · 2012-09-11, 13:50

This project could use a boinc setup. [Not volunteering!]

MrHappy · 2012-09-11, 20:32

Is "FermatFactor=127,3700000e9,4000000e9" the correct worktodo entry for the reservation of: 127 3,700,000,000,000,000 4,000,000,000,000,000 ?
I'm a bit confused with your k, N, and things...

;)

Jatheski · 2012-09-11, 20:45

Yes or

FermatFactor=127,3700e12,4000e12 ;)

RichD · 2012-09-11, 23:55

It looks like this is still available so I would like to reserve it.

Code:

MMFactor=127,2.5e15,2.8e15

And it will take 8 days on my GTX 560.

flashjh · 2012-09-12, 04:15

Quote:

Originally Posted by MrHappy

Is "FermatFactor=127,3700000e9,4000000e9" the correct worktodo entry for the reservation of: 127 3,700,000,000,000,000 4,000,000,000,000,000 ?
I'm a bit confused with your k, N, and things...

;)

Quote:

Originally Posted by Jatheski

Yes or

FermatFactor=127,3700e12,4000e12 ;)

I thought you had to add '2' now?: FermatFactor=129,3700e12,4000e12

Quote:

Originally Posted by Prime95

Here is the next version of mmff.
If you reserved a Fermat range from fermatsearch.org, please retest it with the new executable (but add 2 to the exponent in worktodo.txt). For example, if you tested a k range for F30 (FermatFactor=30,1.5e15,1.6e15) using the previous mmff, Luigi had you reserve a range using N=32. With this executable, retest the range using FermatFactor=32,1.5e15,1.6e15.

rcv · 2012-09-12, 05:12

Quote:

Originally Posted by flashjh

I thought you had to add '2' now?: FermatFactor=129,3700e12,4000e12

Quote:

Originally Posted by prime95

Here is the next version of mmff.
If you reserved a Fermat range from fermatsearch.org, please retest it with the new executable (but add 2 to the exponent in worktodo.txt). For example, if you tested a k range for F30 (FermatFactor=30,1.5e15,1.6e15) using the previous mmff, Luigi had you reserve a range using N=32. With this executable, retest the range using FermatFactor=32,1.5e15,1.6e15.

My understanding is as follows:

MMFactor work.run under mmff-0.20 is OK.
FermatFactor work run under mmff-0.20 should be discarded.

The interpretation of the FermatFactor line has changed:

Code:

 mmff-0.20:  FermatFactor=<N-2>,low,high

Code:

 mmff-0.21:  FermatFactor=N,low,high

If you get new reservations, the N and the k range you receive from Luigi should be plugged directly into the FermatFactor line. Simple. Straightforward. Ergonomic.

If you got reservations while using the old (0.20) version of mmff, you had to tweak the N value, by subtracting 2. To move that reservation to the new (0.21) version of mmff, you must undo your tweak.

LaurV · 2012-09-12, 05:16

"Reservation" means "N range, k range". The "+2" is there already. The correct one is 127....
This is what the site use, and the new software version. This can find factors of Fm with m<=125. Usually N>=m+2.

The older version of the program would have had to reserve the same range as "125....".

So, please do not use 129!

frmky · 2012-09-12, 07:17

Code:

no factor for MM107 from 2^155 to 2^156 (k range: 200T to 300T) [mmff 0.21 mfaktc_barrett160_M107gs]
no factor for MM107 from 2^156 to 2^157 (k range: 0T to 300T) [mmff 0.21 mfaktc_barrett160_M107gs]

2012-09-11, 23:55	#139
RichD Sep 2008 Kansas 3376₁₀ Posts	It looks like this is still available so I would like to reserve it. Code: MMFactor=127,2.5e15,2.8e15 And it will take 8 days on my GTX 560.

2012-09-12, 07:17	#143
frmky Jul 2003 So Cal 2·3⁴·13 Posts	Code: no factor for MM107 from 2^155 to 2^156 (k range: 200T to 300T) [mmff 0.21 mfaktc_barrett160_M107gs] no factor for MM107 from 2^156 to 2^157 (k range: 0T to 300T) [mmff 0.21 mfaktc_barrett160_M107gs]

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Mersenne trial division implementation	mathPuzzles	Math	8	2017-04-21 07:21
trial division over a factor base	Peter Hackman	Factoring	7	2009-10-26 18:27
P95 trial division strategy	SPWorley	Math	8	2009-08-24 23:26
Trial division software for Mersenne	SPWorley	Factoring	7	2009-08-16 00:23
Need GMP trial-division timings	ewmayer	Factoring	7	2008-12-11 22:12

2012-09-11, 13:50	#136
axn Jun 2003 5051₁₀ Posts	This project could use a boinc setup. [Not volunteering!]

2012-09-11, 20:32	#137
MrHappy Dec 2003 Paisley Park & Neverland 5·37 Posts	Is "FermatFactor=127,3700000e9,4000000e9" the correct worktodo entry for the reservation of: 127 3,700,000,000,000,000 4,000,000,000,000,000 ? I'm a bit confused with your k, N, and things... ;)

2012-09-11, 20:45	#138
Jatheski Apr 2012 993438: i1090 2×73 Posts	Yes or FermatFactor=127,3700e12,4000e12 ;)

2012-09-12, 05:16	#142
LaurV Romulan Interpreter Jun 2011 Thailand 7×1,373 Posts	"Reservation" means "N range, k range". The "+2" is there already. The correct one is 127.... This is what the site use, and the new software version. This can find factors of Fm with m<=125. Usually N>=m+2. The older version of the program would have had to reserve the same range as "125....". So, please do not use 129!