S9 and general sieving discussion
 

[QUOTE=MyDogBuster;382092]Max plans on updating them to a newer version. Neither port is loaded yet. The URL's will be updated when Max creates everything.[/QUOTE]


Could you stop loading those candidates ?

How high is your files on s9 sieved ? I started sieve s9 from 2T-3T and it finds factors very fast.

Lennart

edit:p=2058652664447, 126759224 p/sec, 6 factors, 5.9% done, 77 sec/factor

This is in a early stage of the sieving but wait some time then I get more info





Edit 2: p=2500616307971, 126432416 p/sec, 43 factors, 50.1% done, 92 sec/factor

1 hr sieve on 4 core


2029518246373 | 2036*9^1567646+1
2035804376783 | 2036*9^1570436+1
2054968490297 | 2036*9^1823396+1
2055094528639 | 2036*9^1887080+1
2058652680871 | 2036*9^1599632+1
2064362027333 | 2036*9^1538312+1
2078779200013 | 2036*9^1887440+1
2079092756729 | 2036*9^1928030+1
2080020305807 | 2036*9^1774796+1
p=2080020292333, 126512898 p/sec, 10 factors, 8.0% done, ETA 04 Sep 17:21

p=2997275315893, 126492258 p/sec, 82 factors, 99.7% done, ETA 04 Sep 17:21
sr2sieve 1.8.10 stopped: at p=3000000000000 because range is complete.
Found factors for 82 terms in 8424.442 sec. (expected about 78.33)

Continue to 4T

Lennart

p=3992614627921, 125676835 p/sec, 56 factors, 99.3% done, ETA 04 Sep 19:39
sr2sieve 1.8.10 stopped: at p=4000000000000 because range is complete.
Found factors for 56 terms in 7951.724 sec. (expected about 54.21)


countinue to 5T

Lennart

p=4999719406139, 125667391 p/sec, 40 factors, 100.0% done, 198 sec/factor
sr2sieve 1.8.10 stopped: at p=5000000000000 because range is complete.
Found factors for 40 terms in 7946.518 sec. (expected about 41.30)

Continue to 6T

Lennart

sr2sieve 1.8.10 stopped: at p=6000000000000 because range is complete.
Found factors for 30 terms in 7955.801 sec. (expected about 33.29)

I take this up to 12T now

Lennart

Oh, that is bad. I did not realize that we had a sieve file that was so woefully undersieved on S9. It appears sieved to only P=2T (before Lennart started on it).

Everybody please hang tight while we get the servers upgraded and the file sieved to an appropriate depth. Do not do any testing on the file.

Max, it would be best if you delete all of the S9 candidates out of the server so that someone does not accidently start testing it.

I am un-sticking this thread until we are more ready with these efforts.

=11997835778477, 124745672 p/sec, 108 factors, 100.0% done, 445 sec/factor
sr2sieve 1.8.10 stopped: at p=12000000000000 because range is complete.
Found factors for 108 terms in 47939.851 sec. (expected about 122.95)



I will sieve more but I think you can start it now.


Lennart


Edit I am sieving to 21T and it will be done today.

sr2sieve 1.8.10 stopped: at p=15000000000000 because range is complete.
Found factors for 45 terms in 24272.608 sec. (expected about 38.50)

12T-15T


Lennart

=17996350338493, 95705997 p/sec, 26 factors, 99.9% done, 1204 sec/factor
sr2sieve 1.8.10 stopped: at p=18000000000000 because range is complete.
Found factors for 26 terms in 31304.105 sec. (expected about 31.27)

Lennart

p=20996427015413, 89710035 p/sec, 22 factors, 99.9% done, 1518 sec/factor
sr2sieve 1.8.10 stopped: at p=21000000000000 because range is complete.
Found factors for 22 terms in 33332.494 sec. (expected about 26.30)

I am doing 21T-24T

p=25995730049801, 122824544 p/sec, 40 factors, 99.9% done, ETA 06 Sep 13:43
sr2sieve 1.8.10 stopped: at p=26000000000000 because range is complete.
Found factors for 40 terms in 40635.497 sec. (expected about 35.55)

Going to 30T

Lennart

26T-30T complete

Doing up to 34T and shall make some test.

Lennart

p=33994197134077, 121846722 p/sec, 21 factors, 99.9% done, 1560 sec/factor
sr2sieve 1.8.10 stopped: at p=34000000000000 because range is complete.
Found factors for 21 terms in 32796.706 sec. (expected about 20.50)


Lennart

[QUOTE=mdettweiler;382415]Well, everyone, we're in luck: I never actually loaded the candidates into the server yet (got swamped with other work). :smile: I'll wait until we're all sure the sieving is done before loading anything.

On another note: do we want to clear out the stats for port 1300 (I think there was some old junk in there from the previous base 2 drive) before starting the new effort? Likewise for port 1400? I can save off a "snapshot" of the stats for posterity before doing so.

Also, I fully intend to upgrade both the PRPnet servers when I load new work in them.[/QUOTE]


You can start loading s9 now when all factors are removed.

Time we save by sieving more is zero in my point of view.

Lennart

OK, I've removed all the factors for 2T-34T from the S9 sieve file. The p=34T file is attached.

I'll be loading this file into port 1300 shortly, after I've cleared out the old stuff in there and upgraded it to 5.3.2.

[QUOTE=mdettweiler;382436]OK, I've removed all the factors for 2T-34T from the S9 sieve file. The p=34T file is attached.

I'll be loading this file into port 1300 shortly, after I've cleared out the old stuff in there and upgraded it to 5.3.2.[/QUOTE]

I am sieving this base higher and post the factors here.

Lennart

Factors for 34T-40T

Lennart

[QUOTE=gd_barnes;382559]Nice work Max in getting these taken care of. Please also remove the P=34T-40T factors and send an updated file to me.

Lennart, are you done sieving S9?

I am now stickying this thread since it is ready for business.[/QUOTE]


No. It need to be sieved higer. I will soon be done with 40T-45T and i'll start 45T-50T after that.

Lennart

40T-45T done

Lennart

[QUOTE=Lennart;382418]You can start loading s9 now when all factors are removed.

Time we save by sieving more is zero in my point of view.

Lennart[/QUOTE]

[QUOTE=Lennart;382580]No. It need to be sieved higer. I will soon be done with 40T-45T and i'll start 45T-50T after that.

Lennart[/QUOTE]

I'm confused. The time saved by sieving is zero but it needs to be sieved higher? lol

I think that Max would not have loaded the file if it needed to be sieved further. If you are going to continue to sieve, I will remove the file from the server.

[QUOTE=gd_barnes;382614]I'm confused. The time saved by sieving is zero but it needs to be sieved higher? lol

I think that Max would not have loaded the file if it needed to be sieved further. If you are going to continue to sieve, I will remove the file from the server.[/QUOTE]


At the n-level we are at now we are ok but we do not have a optimal sieve for 1.9M.

I would not have load all at once I should have loaded 1.5M-1.6M
and continue to sieve

There is another problem also. what type of CPU shall I use when comparing time ?

If I use my AVX core I think I could stop now but if I use a time from a 3Ghz none AVX I need to sieve more.

Lennart

[QUOTE=Lennart;382619]At the n-level we are at now we are ok but we do not have a optimal sieve for 1.9M.

I would not have load all at once I should have loaded 1.5M-1.6M
and continue to sieve

There is another problem also. what type of CPU shall I use when comparing time ?

If I use my AVX core I think I could stop now but if I use a time from a 3Ghz none AVX I need to sieve more.

Lennart[/QUOTE]

Max has made me aware of the way PrimeGrid/PSP does sieving/testing on things at the same time, which we have not done in the past at NPLB/CRUS. From this point on, I will defer to our searchers with such experience when determining how far to sieve and when to start testing. For this, I'll defer to your knowledge on the topic. What would PrimeGrid do?

I feel that technically what we should do is use the type of machine that will be most used for sieving for a sieving rate and the type of machine that will be most used for testing in the future for a testing rate.

One question for you: As a simple example, let's say that we only loaded n=1.5M-1.6M into the server. Would you still continue to sieve n=1.5M to 2M further or would you only sieve n=1.6M-2M further?

[QUOTE=gd_barnes;382667]Max has made me aware of the way PrimeGrid/PSP does sieving/testing on things at the same time, which we have not done in the past at NPLB/CRUS. From this point on, I will defer to our searchers with such experience when determining how far to sieve and when to start testing. For this, I'll defer to your knowledge on the topic. What would PrimeGrid do?

I feel that technically what we should do is use the type of machine that will be most used for sieving for a sieving rate and the type of machine that will be most used for testing in the future for a testing rate.

One question for you: As a simple example, let's say that we only loaded n=1.5M-1.6M into the server. Would you still continue to sieve n=1.5M to 2M further or would you only sieve n=1.6M-2M further?[/QUOTE]


If I run the tests on prpnet I should leave all candidates in the sievefile because it is very easy to remove factors by prpadmin.

It only takes some seconds so I use that.

Because of the slow computers I see now on s9 (not all) I started the sieving again and I think there is no harm in doing that.

I think this is a good way of sieving specially when it is a conjecture when you don't know where you going to find the primes that will eliminate a k.

When we come to time for tests it is very hard today If I do a test on a number with a modern i7 3.2Ghz no AVX can take 2 hr but on a i7 with AVX it will only take 1 hr.

On a c2q6600 it will take about 4 -5hr



Lennart

I have one core of an AMD X6 1000T on the S9 port currently. Is taking about 7.5 hours to crunch one WU (6 completed in 44:36:55).

What machine to base the sieving level on? I would say the average machine over the life of the port should be the basis. Not my AMD CPU, (non AVX) etc. Maybe I will upgrade to Skylake next year with DDR4 and AVX-512?

[QUOTE=TheCount;382684]I have one core of an AMD X6 1000T on the S9 port currently. Is taking about 7.5 hours to crunch one WU (6 completed in 44:36:55).

What machine to base the sieving level on? I would say the average machine over the life of the port should be the basis. Not my AMD CPU, (non AVX) etc. Maybe I will upgrade to Skylake next year with DDR4 and AVX-512?[/QUOTE]


Most sieving is done on i7 2700 at 3.7Ghz.

It is on about 6 hr/f on that computer.
and I do them in about 4 hr when I llr them on that computer


Lennart

45T-54T completed.


I will do some more because I see tests on the s9 port is not very fast.

Lennart

[QUOTE=Lennart;382703]45T-54T completed.


I will do some more because I see tests on the s9 port is not very fast.

Lennart[/QUOTE]
I should note: take the candidate ages you see reserved to my computers on the server with a grain of salt. :smile: Both of the computers I have running S9 right now only do PRPnet part-time (mostly at night) - they're laptops that go places during the day.

For a more accurate figure, I'm getting about 17,200 seconds = 4.7 hours/test on one of the laptops (Sandy Bridge i7 dualcore - i7-2620M).

I just removed the factors up to P=54T in port 1300.

Some more factors.

Lennart

This is a strange sieve! I would like to get some sec. more but it moves very slow.

62T-70T done

I will now do it up to 80T and see if time increase anything.

Lennart

Factors up to P=70T removed in port 1300.

70T-80T complete and I stop here I am at ¨8 hr now.


Lennart