Twin prime search? - Page 7

MooMoo2 · 2006-05-11, 00:31

Quote:

Originally Posted by pacionet

We are online at:

http://www.twinprimesearch.org

I can set up up mail accounts of the form:

username AT twinprimesearch.org

with automatic redirection on your normal email (ALIAS)

If you want one email me your desired mail and your normal email at :

info AT twinprimesearch.org

(For example, MooooMoo I can replace your email on the site moo.the.cow800 AT gmail.com with reserve AT twinprimesearch.org if you want and you received emails at your normal address moo.the.cow800 AT gmail.com ).

Just to point out some minor errors/update issues on the site:

- the "join" section should be updated (copy/paste the first post of the stickied "project instructions" thread). The main reason for this update is that new users should download pre-sieved ranges, so they don't need NewPGen.

- range 29M-30M (reserved by Zeus) is missing

- All ranges below 30M have been completed, and there is one prime in the 10M-15M range (found by victor)

Other than that, the website is excellent

I'm also looking forward to having the project on Kirk Pearson's DC site.

pacionet · 2006-05-11, 23:11

Ok, victor has confirmed me that email redirecting works. SO if you want some account of the form:

NAME@twinprimesearch.org

tell me

I sent to mr. Pearson the request to appear in his site.

Now I started to work to dinamic version of the site.

pacionet · 2006-05-12, 11:04

I send the link to our site to some search engines.
If you know forums, sites , blog , please try to publicize our site (and forum).

MooMoo2 · 2006-05-18, 23:36

Quote:

Originally Posted by pacionet

Ok, victor has confirmed me that email redirecting works. SO if you want some account of the form:

NAME@twinprimesearch.org

tell me

I sent to mr. Pearson the request to appear in his site.

Now I started to work to dinamic version of the site.

Unfortunately, on the site's newest update, our project is not included.

pacionet · 2006-05-25, 07:56

Kirk Pearson has added our project to his site:

http://distributedcomputing.info

which is the best site about distributed computing.

pacionet · 2006-05-27, 19:45

I found the page of the Hungarians researchers which holds the current record for the largest twin primes:

http://compalg.inf.elte.hu/~wizo/?kutatas&en

I hope we can beat them !

R. Gerbicz · 2006-05-27, 20:31

Quote:

Originally Posted by pacionet

I found the page of the Hungarians researchers which holds the current record for the largest twin primes:

http://compalg.inf.elte.hu/~wizo/?kutatas&en

I hope we can beat them !

Good work! But I'm proud that an Hungarian team holds the world record. Here you can find more records obtained by the project leader Antal Járai: http://compalg.inf.elte.hu/~ajarai/worldr.htm

biwema · 2006-05-27, 23:11

Did someone of you already take a close look at the FFT lengths?

On my P4 (SSE2) the exponent 195000 has a 20k FFT.
Is that the same with non SSE2 machines or do they use 16k FFT?

The last record Twin has the exponent 171960. That is a 16k FFT. That FFT lengt is enough for exponents up to 172500 (I tested with a mantissa of k= 1T).
The probability of a number of that size being prime is larger than 195000.

Hence the expected time of finding a twin is only 54% compared to 195000.

----

I also noticed that testing +1 first is slightly faster on a P4 (besides the chance of finding a fermat divisor).

So I recommend to replace the first line of the range file with
[givenfactoringlimit]:P:0:2:1

running a notmal +1 test with LLR
All Primes are written into the output file.

In that file you cann add the first line
[givenfactoringlimit]:M:0:2:2

The primes that are remaining now are the twins.
So it is more unlike to miss a prime.

----

Calculation Times considering FFT lengths

Again a Table...
Sieving limit: 35T (CPU P4 3.4GHz)

Exp____FFT__Tests___iter___Twinevery__Twintime
172500__16k__4.6M__0.379ms__10.6G___9.5 CPUyears
195000__20k__5.9M__0.490ms__13.5G___17.8 CPUyears
250000__24k__9.7M__0.597ms__22.2G___45.9 CPUyears
280000__28k_12.2M__0.719ms__27.8G___77.9 CPUyears
333000__32k_17.3M__0.800ms__39.5G___146 CPUyears

MooMoo2 · 2006-05-28, 07:41

biwema,

I disagree that testing +1 first is slightly faster on a P4. While finding a recent prime, this is what I got:

168367377*2^195000-1 is prime! Time : 95.454 sec.
168367377*2^195000+1 is not prime. Proth RES64: 2A409B23B3D7987E Time: 95.728 sec.

edit: I forgot to mention that it's almost impossible to find Fermat divisors for large k.

pacionet · 2006-05-28, 13:55

Also on my Pentium 4, 3.2 Ghz testing -1 k's is faster:

186002877*2^195000-1 is prime! Time : 104.031 sec.
186002877*2^195000+1 is not prime. Proth RES64: 29FD50C0848E4FEE Time: 104.398 sec.

biwema · 2006-05-28, 16:26

I see.

I did not a whole test, but just some 10000 iterations on different fft length. I set the output every 2000 iterations. Here +1 was 0.2% - 0.5% faster. Maybe it is dependant on the memory and cache. I also noticed that the iteration times increases 1% every 8000 - 12000 iterations. I don't know why.

2006-05-25, 07:56	#71
pacionet Oct 2005 Italy 3·113 Posts	Kirk Pearson has added our project to his site: http://distributedcomputing.info which is the best site about distributed computing. Last fiddled with by pacionet on 2006-05-25 at 08:02

2006-05-28, 07:41	#75
MooMoo2 "Michael Kwok" Mar 2006 1,181 Posts	biwema, I disagree that testing +1 first is slightly faster on a P4. While finding a recent prime, this is what I got: 1683673772^195000-1 is prime! Time : 95.454 sec. 1683673772^195000+1 is not prime. Proth RES64: 2A409B23B3D7987E Time: 95.728 sec. edit: I forgot to mention that it's almost impossible to find Fermat divisors for large k. Last fiddled with by MooMoo2 on 2006-05-28 at 07:44

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
A Twin Prime Search Method	jzakiya	jzakiya	12	2018-12-11 21:09
Highest Prime is also a twin prime... NOT	hydeer	Lone Mersenne Hunters	9	2018-04-03 22:54
Twin Prime Days, Prime Day Clusters	cuBerBruce	Puzzles	3	2014-12-01 18:15
Top-10 twin prime found!	gd_barnes	Riesel Prime Search	6	2007-10-12 18:42
Questions about twin prime search	jasong	Twin Prime Search	14	2006-12-14 23:16

2006-05-11, 23:11	#68
pacionet Oct 2005 Italy 3×113 Posts	Ok, victor has confirmed me that email redirecting works. SO if you want some account of the form: NAME@twinprimesearch.org tell me I sent to mr. Pearson the request to appear in his site. Now I started to work to dinamic version of the site.

2006-05-12, 11:04	#69
pacionet Oct 2005 Italy 3·113 Posts	I send the link to our site to some search engines. If you know forums, sites , blog , please try to publicize our site (and forum).

2006-05-27, 19:45	#72
pacionet Oct 2005 Italy 3·113 Posts	I found the page of the Hungarians researchers which holds the current record for the largest twin primes: http://compalg.inf.elte.hu/~wizo/?kutatas&en I hope we can beat them !

2006-05-27, 23:11	#74
biwema Mar 2004 3·127 Posts	Did someone of you already take a close look at the FFT lengths? On my P4 (SSE2) the exponent 195000 has a 20k FFT. Is that the same with non SSE2 machines or do they use 16k FFT? The last record Twin has the exponent 171960. That is a 16k FFT. That FFT lengt is enough for exponents up to 172500 (I tested with a mantissa of k= 1T). The probability of a number of that size being prime is larger than 195000. Hence the expected time of finding a twin is only 54% compared to 195000. ---- I also noticed that testing +1 first is slightly faster on a P4 (besides the chance of finding a fermat divisor). So I recommend to replace the first line of the range file with [givenfactoringlimit]:P:0:2:1 running a notmal +1 test with LLR All Primes are written into the output file. In that file you cann add the first line [givenfactoringlimit]:M:0:2:2 The primes that are remaining now are the twins. So it is more unlike to miss a prime. ---- Calculation Times considering FFT lengths Again a Table... Sieving limit: 35T (CPU P4 3.4GHz) Exp____FFT__Tests___iter___Twinevery__Twintime 172500__16k__4.6M__0.379ms__10.6G___9.5 CPUyears 195000__20k__5.9M__0.490ms__13.5G___17.8 CPUyears 250000__24k__9.7M__0.597ms__22.2G___45.9 CPUyears 280000__28k_12.2M__0.719ms__27.8G___77.9 CPUyears 333000__32k_17.3M__0.800ms__39.5G___146 CPUyears

2006-05-28, 13:55	#76
pacionet Oct 2005 Italy 153₁₆ Posts	Also on my Pentium 4, 3.2 Ghz testing -1 k's is faster: 1860028772^195000-1 is prime! Time : 104.031 sec. 1860028772^195000+1 is not prime. Proth RES64: 29FD50C0848E4FEE Time: 104.398 sec.

2006-05-28, 16:26	#77
biwema Mar 2004 381₁₀ Posts	I see. I did not a whole test, but just some 10000 iterations on different fft length. I set the output every 2000 iterations. Here +1 was 0.2% - 0.5% faster. Maybe it is dependant on the memory and cache. I also noticed that the iteration times increases 1% every 8000 - 12000 iterations. I don't know why.