mersenneforum.org  

Go Back   mersenneforum.org > Prime Search Projects > Conjectures 'R Us

Reply
 
Thread Tools
Old 2016-06-16, 15:36   #1376
rebirther
 
rebirther's Avatar
 
Sep 2011
Germany

2×32×151 Posts
Default

Quote:
Originally Posted by rebirther View Post
R52 tested to n=100k (50-100k)

28 primes found, 100 remain

12827*52^53709-1
69111*52^56234-1
82119*52^58295-1
48759*52^58956-1
33071*52^59206-1
56603*52^59512-1
12423*52^59835-1
16308*52^60156-1
39717*52^60574-1
16007*52^61277-1
16748*52^61624-1
12719*52^66440-1
54693*52^66576-1
9624*52^68479-1
47063*52^69168-1
27767*52^71877-1
62463*52^73310-1
46673*52^80919-1
47913*52^81807-1
58407*52^81857-1
71763*52^84127-1
12401*52^84322-1
55119*52^90896-1
39168*52^91095-1
70703*52^93799-1
11394*52^94360-1
8693*52^95515-1
3870*52^99193-1

Results emailed, Base released
@Gary:
I have one more prime in my list which I have forgot to add:

72344*52^74348-1

I will resend you the primefile for this base. There are 99 remain.
rebirther is offline   Reply With Quote
Old 2016-06-16, 17:00   #1377
rebirther
 
rebirther's Avatar
 
Sep 2011
Germany

2·32·151 Posts
Default

Reserving R94 to n=1M (586.7k-1M) for BOINC
rebirther is offline   Reply With Quote
Old 2016-06-17, 00:36   #1378
paleseptember
 
paleseptember's Avatar
 
Jun 2008
Wollongong, .au

3·61 Posts
Default S49

S49 has two remaining k at n=600K. I'll try to extend that out to n=1M.

Since there are only two k remaining, how far should I go with sieving? When removing a factor take 50%? 80%? of the time to complete a primality test at n= ~1M?
paleseptember is offline   Reply With Quote
Old 2016-06-17, 06:17   #1379
pepi37
 
pepi37's Avatar
 
Dec 2011
After milion nines:)

25758 Posts
Default

Quote:
Originally Posted by paleseptember View Post
S49 has two remaining k at n=600K. I'll try to extend that out to n=1M.

Since there are only two k remaining, how far should I go with sieving? When removing a factor take 50%? 80%? of the time to complete a primality test at n= ~1M?
Go high as you can. Conjuncture is weird thing, maybe your prime is inside that range, maybe not. And every single candidate that sieve remove, dont need to be tested. I sow on this forum that recommended is 80% .
On the other side, if you find prime at beginning of your search then you will have many candidates removed...
Decision is yours :)
pepi37 is offline   Reply With Quote
Old 2016-06-17, 15:21   #1380
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2×3×19×41 Posts
Default

Quote:
Originally Posted by paleseptember View Post
S49 has two remaining k at n=600K. I'll try to extend that out to n=1M.

Since there are only two k remaining, how far should I go with sieving? When removing a factor take 50%? 80%? of the time to complete a primality test at n= ~1M?
If you knew you were going to test the entire file, you'd optimally sieve until factor-found rate matched the testing time for a candidate 70% of the way from n-min to n-max (~880k in your case). That's roughly 75% of testing time for a candidate at 1M, using (880/1000)^2 to estimate relative testing times.

However, you are not planning to test the entire file- if you find a prime, half the remaining tests will not be run. You could use the heuristic for chance of finding a prime, and solve for optimal sieve time; or you could take a wild-ass-guess and sieve to a point somewhat below 75% of 1M testing time and call it good. My WAG is 10% chance of prime in the entire file, so I'd sieve to about 70% of the 1M testing time; remember that optimal sieving *barely* matters in overall project length; you could get away with 50% or 100% and it won't make very much difference.
VBCurtis is offline   Reply With Quote
Old 2016-06-17, 17:00   #1381
pepi37
 
pepi37's Avatar
 
Dec 2011
After milion nines:)

5×281 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
..... remember that optimal sieving *barely* matters in overall project length; you could get away with 50% or 100% and it won't make very much difference.
If I may say: that logic is OK if you have many CPU cores. But if you have limited number of CPU cores , than every candidate removed by sieve is at least one hour less on fast AVX core
pepi37 is offline   Reply With Quote
Old 2016-06-17, 17:41   #1382
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT/BST)

2×2,909 Posts
Default

Quote:
Originally Posted by pepi37 View Post
If I may say: that logic is OK if you have many CPU cores. But if you have limited number of CPU cores , than every candidate removed by sieve is at least one hour less on fast AVX core
If you have fast and slow cores then yes you could change things around.
henryzz is offline   Reply With Quote
Old 2016-06-17, 22:24   #1383
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2×3×19×41 Posts
Default

Quote:
Originally Posted by pepi37 View Post
If I may say: that logic is OK if you have many CPU cores. But if you have limited number of CPU cores , than every candidate removed by sieve is at least one hour less on fast AVX core
Huh? How does the number of cores have anything to do with the computation of optimal sieve depth? My advice is based on minimum project length- you can use just as many cores to sieve as you do to LLR, or you can use one core for the entire thing, or whatever you want; the project length is what we're trying to minimize, measured in computational effort rather than wall-clock time.

Are you saying a factored candidate will save more than one hour on your AVX core? Then you should sieve until it takes the sieve an hour or so to factor a candidate. I still don't get what you are trying to say.
VBCurtis is offline   Reply With Quote
Old 2016-06-18, 06:36   #1384
Puzzle-Peter
 
Puzzle-Peter's Avatar
 
Jun 2009

683 Posts
Default

As somebody stated before, optimum sieve depth is impossible to determine as we hope we don't need to test the whole candidate file. I've had files that I sieved from n=1M to n=2M only to find a prime at n=1.05M and all the sieving I did was basically wasted.

That's why I do some basic sieving, test a few candidates at different n-levels to get timing information and then switch from sieving to LLR and back and forth, depending on what is faster.
Puzzle-Peter is offline   Reply With Quote
Old 2016-06-18, 12:36   #1385
pepi37
 
pepi37's Avatar
 
Dec 2011
After milion nines:)

5×281 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
....
Are you saying a factored candidate will save more than one hour on your AVX core? Then you should sieve until it takes the sieve an hour or so to factor a candidate. I still don't get what you are trying to say.

Yes, that is what I try to say. We all hope that prime will be at beginning of the file we process. But what if it on the end, or even there is no prime in our range. In that case, every candidate removed with sieve is less to process.
pepi37 is offline   Reply With Quote
Old 2016-06-19, 13:11   #1386
rebirther
 
rebirther's Avatar
 
Sep 2011
Germany

1010100111102 Posts
Default

Quote:
Originally Posted by rebirther View Post
Reserving S55 to n=500k (250-500k) for BOINC
Reserving to n=1M
rebirther is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Bases 501-1030 reservations/statuses/primes KEP Conjectures 'R Us 3828 2021-02-27 09:19
Bases 251-500 reservations/statuses/primes gd_barnes Conjectures 'R Us 2257 2021-02-27 09:18
Bases 101-250 reservations/statuses/primes gd_barnes Conjectures 'R Us 890 2021-02-24 18:35
Riesel base 3 reservations/statuses/primes KEP Conjectures 'R Us 1081 2021-02-23 16:13
Bases 4-32 reservations/statuses/primes gd_barnes Conjectures 'R Us 1431 2021-02-18 16:05

All times are UTC. The time now is 19:33.

Mon Mar 1 19:33:39 UTC 2021 up 88 days, 15:44, 1 user, load averages: 1.46, 1.54, 1.79

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

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.