mersenneforum.org Can someone explain these P-1 Relative Primes variances?
 Register FAQ Search Today's Posts Mark Forums Read

 2020-01-12, 05:11 #1 petrw1 1976 Toyota Corona years forever!     "Wayne" Nov 2006 Saskatchewan, Canada 428810 Posts Can someone explain these P-1 Relative Primes variances? I have 3 PC's all running P-1 with the same bounds with exponents in the same range. However, all 3 are process a different number of Relative Primes; RPs. (And there is a 4th and 5th quite close in exponents ALSO with different RPs) All are running: PMINUS1=N/A,1,2,41xxxxxx,-1,750000,15000000,74 1. 41,8xx,xxx; 336 RPs; 6,500 RAM; 4 Workers X 1 Core; i5-3570; v29.4b5 2. 41,8xx,xxx; 192 RPs; 4,000 RAM; 2 X 2; i5-3570K; v29.4b7 3. 41,8xx,xxx; 480 RPs; 14,000 RAM; 4 X 1; i5-3570; v29.4b5 . . . . . 4. 41,3xx,xxx; 960 RPs; 20,000 RAM; 8 X 1; i7-7820X; v29.5b3 5. 42,3xx,xxx; 480 RPs; 11,000 RAM; 4 X 1; i5-2500; v29.4b7 Is it simply a function of Cores and RAM? Or is it also a function of CPU Specs? Or Version of Prime95? Do the RPs affect GhzDays given? Odds of finding a factor? Thanks Last fiddled with by petrw1 on 2020-01-12 at 05:13 Reason: Last 2 ?s
 2020-01-12, 05:53 #2 Prime95 P90 years forever!     Aug 2002 Yeehaw, FL 22·1,709 Posts RPs are a function of how much RAM prime95 thinks this a worker can use and stay under the memory limit your gave. RPs do not affect CPU credit. RPs do not affect chance of finding a factor.
2020-01-12, 06:08   #3
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

382410 Posts

Quote:
 Originally Posted by petrw1 Is it simply a function of Cores and RAM? Or is it also a function of CPU Specs? Or Version of Prime95? Do the RPs affect GhzDays given? Odds of finding a factor? Thanks
As I understand it, for given bounds and exponent, #RP is a function of ram available and software design.
#RP will affect run time; more #RP is more efficient.

Bounds and prior factoring affect factoring odds for the P-1 work. https://www.mersenneforum.org/showpo...4&postcount=17

Just curious, why are you still running old V29.4? V29.8 has been out for almost a year.

Last fiddled with by kriesel on 2020-01-12 at 06:13

2020-01-12, 16:17   #4
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

26×67 Posts

Quote:
 RPs are a function of how much RAM prime95 thinks this a worker can use and stay under the memory limit your gave.
Code:
[Jan 11 14:44] Available memory is 5265MB.
[Jan 11 14:44] Using 5263MB of memory.  Processing 296 relative primes (0 of 960 already processed).
I understand that the 296 above is chosen based on the available RAM; 5265MB in this case.
But my question is what determines the second number in bold above?
Again I have 5 computers working on very similar P1 assignments and 4 of them have a different total RP number.

Quote:
 Bounds and prior factoring affect factoring odds for the P-1 work.
https://www.mersenne.ca/prob.php has a function to calculate P1 odds taking into account prior TF.
Is there a similar function somewhere that will calculate TF odds for the next bit level taking into account prior P-1 (and ECM)?

Quote:
 why are you still running old V29.4? V29.8 has been out for almost a year.
Laziness; and I didn't notice anything in the new version that affects P1 but I probably missed that it actually does.

Thanks both.

2020-01-12, 16:53   #5
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

22·1,709 Posts

Quote:
 Originally Posted by petrw1 Code: [Jan 11 14:44] Available memory is 5265MB. [Jan 11 14:44] Using 5263MB of memory. Processing 296 relative primes (0 of 960 already processed). I understand that the 296 above is chosen based on the available RAM; 5265MB in this case. But my question is what determines the second number in bold above?
Available memory affects both the 296 and 960 numbers.

2020-01-12, 17:03   #6
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

24·239 Posts

Quote:
 Originally Posted by petrw1 [CODE][Jan 11 14:44] https://www.mersenne.ca/prob.php has a function to calculate P1 odds taking into account prior TF.
For P-1, apply it twice; once for the bounds already done, and once for the bounds planned. Subtract the earlier from the later.
Or see the source code in prime95/mprime or cudapm1. It's not simple.

2020-01-12, 19:44   #7
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

26·67 Posts

Quote:
 Originally Posted by kriesel For P-1, apply it twice; once for the bounds already done, and once for the bounds planned. Subtract the earlier from the later. Or see the source code in prime95/mprime or cudapm1. It's not simple.
Yes, I understand and do that....

By the way, for some reason I am doing reasonably better than this.
For the last 2 years I have been re-doing P-1 in 40M-59M where B1=B2.
Quite consistently the difference between the current B1=B2 (about 650,000/650,000 … about 1.8%)
and the new B1/B2 (usually 1,000,000/20,000,000 … about 4.4%)
as you can calculate is about 2.6%.
However I am quite consistently getting very, very close to 3%.
And that is with a decent sample size of about 25,000 runs.

My question, though, is how can I calculate the expected TF success rate where a little or a lot of P1/ECM has been done?

According to "The Math" the expected success rate for Bit x is about 1/(x+1).
https://www.mersenne.org/various/math.php
So if I am TF'ing at somewhere in the 72 to 75 bit ranges I should expect a success rate of about 1.35%.
But again with a large sample size (about 50,000) my average is consistently very, very close to 1%.
All of these TF runs are for exponents that have had some P1 done.

Thanks.

2020-01-12, 19:48   #8
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

10C016 Posts

Quote:
 Originally Posted by Prime95 RPs do not affect CPU credit.
Code:
1. 41,8xx,xxx; 336 RPs; 6,500 RAM; 4 Workers X 1 Core; i5-3570; v29.4b5
2. 41,8xx,xxx; 192 RPs; 4,000 RAM; 2 X 2; i5-3570K; v29.4b7
3. 41,8xx,xxx; 480 RPs; 14,000 RAM; 4 X 1; i5-3570; v29.4b5
1. 3.3783 Ghz Days
2. 3.2622 Ghz Days
3. 3.3783 Ghz Days

Is that due to the Version of PrimeNet?

2020-01-12, 22:17   #9
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

1AB416 Posts

Quote:
 Originally Posted by petrw1 Is that due to the Version of PrimeNet?
I assume you meant prime95, not primenet.

Prime95 may have chosen to use a different FFT size or different bounds. That would affect CPU credit

2020-01-12, 22:20   #10
masser

Jul 2003

53216 Posts

Quote:
 Originally Posted by petrw1 Code: 1. 41,8xx,xxx; 336 RPs; 6,500 RAM; 4 Workers X 1 Core; i5-3570; v29.4b5 2. 41,8xx,xxx; 192 RPs; 4,000 RAM; 2 X 2; i5-3570K; v29.4b7 3. 41,8xx,xxx; 480 RPs; 14,000 RAM; 4 X 1; i5-3570; v29.4b5 1. 3.3783 Ghz Days 2. 3.2622 Ghz Days 3. 3.3783 Ghz Days Is that due to the Version of PrimeNet?
No. My guess is that the low ram instance did not do the Brent-Sumaya extension; I've noticed that I get a little more credit with the E=12 results.

Last fiddled with by masser on 2020-01-12 at 22:21

2020-01-12, 22:43   #11
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

103008 Posts

Quote:
 Originally Posted by Prime95 I assume you meant prime95, not primenet. Prime95 may have chosen to use a different FFT size or different bounds. That would affect CPU credit
FFTs in order from post#1 are:
2304K
2240K
2304K
2160K
2240K

Seems the smaller the FFT chosen the lower the credit given.
Makes sense; thanks.

 Similar Threads Thread Thread Starter Forum Replies Last Post wildrabbitt Homework Help 13 2019-09-29 16:23 petrw1 Factoring 4 2018-11-07 20:20 Fred Software 5 2016-04-01 18:15 tha Software 14 2015-10-30 01:23 timbit Information & Answers 0 2009-03-13 18:45

All times are UTC. The time now is 05:42.

Mon Jun 1 05:42:35 UTC 2020 up 68 days, 3:15, 1 user, load averages: 1.16, 1.30, 1.31

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.