Private TPS at n=666777

[Svenie25]

Hello folks.

Yesterday evenning I started a TPS at n=666777 for my one. At the moment I use only one core for it, because I had to finish a manualreserved range for NPLB.
So I started sieving up to a k-rate of 600 seconds for k=2-1M. 600 seconds because a LLR-test of 1M takes so long.
Once this sieve is finished, I will test the left k´s.
In future, then NPLB is finished, I will use one core for sieving and one for LLR.
I will give regularly a statusreport. If someone is in need for the sieved files or so, please post here or pm me for them.

But there is another question left. If I find a prime, twin or not, and I´ll report it at top5000, have I to choose a project? Or is it sufficiently then I choose LLR and NewPGen?

Have a nice day!

[Svenie25]

First sieving is done. k=2-1M left 942 of 999999 candidates left. It is sieved up to p=9.8 trillion. Now starting to llr them.

[gd_barnes]

The effort will likely take about 1000 CPU years to find a single twin at current computer speeds. If you have 942 candidates to test for an n=1M range, you'll need to sieve a range of nearly k=8M to average just one Riesel prime.

It's far more efficient to sieve a much larger k-range at once. If you're going to complete the testing for it, I would suggest sieving a k=10M or 100M range each time.

Although the admins would have to answer for sure, you can choose to put a project in your prover code or not. It's up to you. I think someone else has been putting TPS in their prover code for a personal twin effort at n=450K.


Good luck!

[Svenie25]

Originally Posted by gd_barnes

The effort will likely take about 1000 CPU years to find a single twin at current computer speeds. If you have 942 candidates to test for an n=1M range, you'll need to sieve a range of nearly k=8M to average just one Riesel prime.

It's far more efficient to sieve a much larger k-range at once. If you're going to complete the testing for it, I would suggest sieving a k=10M or 100M range each time.

Although the admins would have to answer for sure, you can choose to put a project in your prover code or not. It's up to you. I think someone else has been putting TPS in their prover code for a personal twin effort at n=450K.


Good luck!

1000 years is an exciting number. But, who knows. Maybe there is a twin at a small k? ;)
About the sieving I was thinking about a larger range, too. So my computer has enough to do over the weekend, when I can´t access to it.
Thanks a lot for the hints. I, or better my computer will do its best.

[Svenie25]

After another good hint from biwema in the other thread, I will sieve the next part of work from k=2M-1000M and then go on in steps of 1000M.

[Svenie25]

Did a "reset". ;)

Started again sieving from k=2 up to this time k=1000M. Sieve will be stopped at k-rate of 645 seconds.

Atm it is at p=2.7 trillion and k-rate of 0.2 seconds.

[gd_barnes]

MUCH better but it is still a large amount of work! Regardless, it will take longer before you can start primality testing but you'll save a lot of time assuming that you will ultimately LLR test the entire range.

Since you had 942 candidates remaining for an n=1M range, with the greater efficiency of sieving, for the n=1G (1000M) range (1000x as big), you will have less than 1000 times as many candidates remaining. On an optimum sieve depth of P=500T (an educated guess), I did a rough calculation and came up with 833000 remaining.

Assuming 645 secs. per test, it will likely take ~17 CPU years to LLR 833000 candidates at n=666777, but you should end up with ~109 Riesel primes from it. From those Riesels, you should have about a 1 in 71 chance of one of them being a Proth and hence a twin.

Each test will have ~1 in 7668 chance of being prime.

If you have 4 quads to dedicate to it full time, you can complete this in slightly over one year! I doubt you'll need to go up in sieving steps of 1000M because it will challenge you greatly to finish the first range. That is unless you plan to make a DC effort out of it -or- you have a lot of machines and a lot of patience! I can promise you, finishing that first range will get very boring! I've been there.


Gary

[Svenie25]

Some Core i7´s would be very cool. ;)

Sievining is going on. p=14.1 trillion, k-rate at 1.2 seconds, 907K canditates left.

[cipher]

Svenie25

I don't want to be rude, but if you do have resources available. Why not help out current TPS search, i see you been helping with sieving. There is no need to start another "n" We would be better off finishing the current TPS before jumping to a new "n"

[gd_barnes]

Originally Posted by Svenie25

Some Core i7´s would be very cool. ;)

Sievining is going on. p=14.1 trillion, k-rate at 1.2 seconds, 907K canditates left.

OK, I speculate that your optimum sieve depth will be a little bit less than:

14.1T * 645 / 1.2 = 7.58P or 7580T or 7.58*10^15

I will guess that it will be around 7P or 7000T.

Man, my educated guess on the optimum sieve depth sucked but the math on the other stuff should be correct. It won't change the estimated # of primes or the odds of a twin but it will probably shorten your total testing time by ~3-5%. Assuming so, it will only take 16 vs. 17 CPU years. :-)

One thing I forgot to mention: The 16 CPU year estimate does not include sieve time. It appears right now that it will take you around 6-9 CPU months of sieving before you can start LLRing; that is if you want to sieve to the optimal depth for the entire job. For "regular" prime searching, sieving is generally about 5-7% of the entire task. For TPS on a single-n, it's usually a little less; more like 2-4%. So 6-9 CPU months vs. 16 CPU years is about right.

Yep, throw a few I7's at this task and you'll be home free! :-)

[Svenie25]

Originally Posted by cipher

Svenie25

I don't want to be rude, but if you do have resources available. Why not help out current TPS search, i see you been helping with sieving. There is no need to start another "n" We would be better off finishing the current TPS before jumping to a new "n"

Call it a personal challenge. ;) The sieving for Primegrid is a little bit depressing. It looks like there is nobody interested in searching this twin. In the last time, only a few people crunched WUs for TPS. And nobody is asking, if they could sieve for more WUs. Looks like, if they can´t find a top5000 prime, they doesn´t do it.

Gary, how do you calculate such things, like the chances of finding a prime?

Sieving is going very well. p is at 60T and there are 825K candidates left.