GPU72 / MISFIT use for 100M digit range? - Page 6

chalsall · 2013-03-02, 18:12

Quote:

Originally Posted by swl551

And why would the credits be so low if the exponents are so large? assignment: exp=332410xxx bit_min=72 bit_max=73 (2.88 GHz-days)

It's actually less expensive to TF a canddiate the higher you go.

LaurV · 2013-03-03, 07:12

Chris explained perfectly the first part (why the interest for 332M). For some of us, we contribute factoring in 332M because is the only range where you still can find lots of factors, and in the same time feel that you do some important job.

The second part (why the lower credit) is related to the special form of the factors: any factor q of a mersenne Mp=2^p-1 has the form q=2*k*p+1 for some natural k=(q-1)/(2*p). Higher p means less number of k's between 2^b and 2^b+1. That is, less candidates to test if they are factors or not. Less candidates, less work, therefore less credit. That is why we can TF 60M to b=73 bits, but M1277 was only TF-ed to 60 bits.

Generally: when you increase the bitlevel by 1, the amount of work (therefore the credit) doubles. When you double the exponent, the amount of work halves.

There would be the same effort (and credit) to TF M1277 to 61 bits (about 120 GhzD) as to TF M65000xxx from 75 to 76 bits, and that is 8 times more than TF-ing M65000xxx to 73 bits.

As 332M is about 5 times higher than 65M, the credit to TF M332M to the same bitlevel (say 73 bits) is 5 times lower (and the mfaktc needs 10 minutes, instead of 50 minutes). That is normal.

Uncwilly · 2013-03-29, 23:00

I felt ashamed that I have to beg GPUto72 to give me more assignments. I am flushing out the last of my primenet assignments and the GPUto72 results have not yet started rolling in in strength.

Aramis Wyler · 2013-03-30, 00:44

Are you doing them breadth first? I went to do a couple the other day (to 80) because someone had mentioned that bitdepth in a forum, and they were just over a day each.

Uncwilly · 2013-03-30, 01:05

Quote:

Originally Posted by Aramis Wyler

Are you doing them breadth first? I went to do a couple the other day (to 80) because someone had mentioned that bitdepth in a forum, and they were just over a day each.

My thrust has been doing the earliest ones to 74 or 76 depending upon the machine (all cpu's, no gpu's). On my borged machines, I stocl them up for about a month every 2 weeks or so.

kracker · 2013-03-30, 02:59

Quote:

Originally Posted by Uncwilly

My thrust has been doing the earliest ones to 74 or 76 depending upon the machine (all cpu's, no gpu's). On my borged machines, I stocl them up for about a month every 2 weeks or so.

That high on cpu's...? imho, it is useless I think....

Uncwilly · 2013-03-30, 03:58

Quote:

Originally Posted by kracker

That high on cpu's...? imho, it is useless I think....

It is my niche. If I had GPU's I would use them too. I have found 14 factors in the last year, each saves ~9,900 GHz Days.
(BTW, I don't dare run LL's on my borged boxen, if one of those hit a prime, the impeller would be impinged upon by biological off-cast.)

kracker · 2013-03-30, 04:39

Quote:

Originally Posted by Uncwilly

It is my niche. If I had GPU's I would use them too. I have found 14 factors in the last year, each saves ~9,900 GHz Days.
(BTW, I don't dare run LL's on my borged boxen, if one of those hit a prime, the impeller would be impinged upon by biological off-cast.)

No, what I mean is they probably would be better off doing other work unless you really like tf'ing on your cpu's...

for example, here is a chart for a Intel 3570k, which is at $220 now, while a nVidia GTX 660 does about ~230 GHz a day, in what benchmarks I see..

bcp19 · 2013-03-30, 14:09

Quote:

Originally Posted by Uncwilly

It is my niche. If I had GPU's I would use them too. I have found 14 factors in the last year, each saves ~9,900 GHz Days.
(BTW, I don't dare run LL's on my borged boxen, if one of those hit a prime, the impeller would be impinged upon by biological off-cast.)

I'm a bit amazed here... I seem to be the only person who got the joke. I do conclude you are making a reference to the defecation contacting a rotary oscillator, hmm?

kladner · 2013-03-31, 02:45

Quote:

Originally Posted by bcp19

I'm a bit amazed here... I seem to be the only person who got the joke. I do conclude you are making a reference to the defecation contacting a rotary oscillator, hmm?

I confess that I did not figure it out until you restated it.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
332.2M - 333.9M (aka 100M digit range)	Uncwilly	LMH > 100M	684	2018-07-01 10:52
I want a 100M digit Mersenne that....	JuanTutors	PrimeNet	8	2012-12-06 13:47
How far along are you in your 100M digit LL test?	JuanTutors	Lounge	6	2012-02-21 07:36
100M-digit n/k pairs	__HRB__	Riesel Prime Search	0	2010-05-22 01:17
100M digit prime	Unregistered	Information & Answers	10	2010-03-24 20:16

2013-03-03, 07:12	#57
LaurV Romulan Interpreter Jun 2011 Thailand 2×5×31² Posts	Chris explained perfectly the first part (why the interest for 332M). For some of us, we contribute factoring in 332M because is the only range where you still can find lots of factors, and in the same time feel that you do some important job. The second part (why the lower credit) is related to the special form of the factors: any factor q of a mersenne Mp=2^p-1 has the form q=2kp+1 for some natural k=(q-1)/(2p). Higher p means less number of k's between 2^b and 2^b+1. That is, less candidates to test if they are factors or not. Less candidates, less work, therefore less credit. That is why we can TF 60M to b=73 bits, but M1277 was only TF-ed to 60 bits. Generally: when you increase the bitlevel by 1, the amount of work (therefore the credit) doubles. When you double the exponent, the amount of work halves*. There would be the same effort (and credit) to TF M1277 to 61 bits (about 120 GhzD) as to TF M65000xxx from 75 to 76 bits, and that is 8 times more than TF-ing M65000xxx to 73 bits. As 332M is about 5 times higher than 65M, the credit to TF M332M to the same bitlevel (say 73 bits) is 5 times lower (and the mfaktc needs 10 minutes, instead of 50 minutes). That is normal.

2013-03-29, 23:00	#58
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 2³·1,223 Posts	I felt ashamed that I have to beg GPUto72 to give me more assignments. I am flushing out the last of my primenet assignments and the GPUto72 results have not yet started rolling in in strength.

2013-03-30, 00:44	#59
Aramis Wyler "Bill Staffen" Jan 2013 Pittsburgh, PA, USA 2³·53 Posts	Are you doing them breadth first? I went to do a couple the other day (to 80) because someone had mentioned that bitdepth in a forum, and they were just over a day each.