mersenneforum.org - Bases 6-32 reservations/statuses/primes

mersenneforum.org (https://www.mersenneforum.org/index.php)

- Conjectures 'R Us (https://www.mersenneforum.org/forumdisplay.php?f=81)

- - Bases 6-32 reservations/statuses/primes (https://www.mersenneforum.org/showthread.php?t=9740)

Riesel base 22 is complete to n=300K. Nothing to report. Now unreserved.

Reserving Riesel base 24 (all 130 k's) from 33.9K up. I'll sieve to n=200K. I'll probably stop somewhere from 50K to 100K depending on when I get tired of it. :smile: (doing some preliminary sieving now, will start it fully after my Sierp base 16 range is done, which is currently at a min N of 168.9K) If anybody wants to help, I'd be open to it. :smile: It's quite a bit of k's, but I can at least get the ball rolling.
Anybody got a suggestion for a sieve depth? Note that I'm using a 32-bit OS and an Athlon. Or just a general tip like how many factors/sec compared to the time to run a candidate some distance into the range, (IIRC the sieve removal rate should be the prime-test of a candidate two thirds of the way through; is that right?) or something.

Sierp base 31 is at n=21K; 41 primes found for n=20K-21K. 1236 k's remain. Continuing...

[quote=Mini-Geek;195342]Reserving Riesel base 24 (all 130 k's) from 33.9K up. I'll sieve to n=200K. I'll probably stop somewhere from 50K to 100K depending on when I get tired of it. :smile: (doing some preliminary sieving now, will start it fully after my Sierp base 16 range is done, which is currently at a min N of 168.9K) If anybody wants to help, I'd be open to it. :smile: It's quite a bit of k's, but I can at least get the ball rolling.
Anybody got a suggestion for a sieve depth? Note that I'm using a 32-bit OS and an Athlon. Or just a general tip like how many factors/sec compared to the time to run a candidate some distance into the range, (IIRC the sieve removal rate should be the prime-test of a candidate two thirds of the way through; is that right?) or something.[/quote]

If you wish to sieve the entire file to the same depth, for the most efficient depth for any effort that will likely complete in < 1-2 years, you should generally PFGW a candidate that is 70% of the n-range (on the machine in which you intend to do testing) and sieve until your removal rate equals approximately that. 2/3rds of the n-range would work well also for the conjectures since the 70% rule is if you were going to test all k's for the entire file, which is not the case for the conjectures.

Technically, it's somewhat more efficient for a range where the highest n-value is > 2X the lowest n-value to sieve the entire file, break off an n-range, test it, and sieve the rest of the file to the optimum depth for just that range (after removing k's with primes). But for only a ~3X ratio here, i.e. 100K/33.9K, with tests that are not in the top-5000 range I wouldn't bother with it. The CPU time savings doesn't quite justify the personal involvement. IMHO, it's only as the tests get into the top-5000 range where the CPU time savings makes it worth it for a ~3X ratio.

A 32-bit O.S. will sieve at about half the speed of a 64-bit one. For that reason, your optimum sieve depth will be half of what it would be otherwise. But if your testing (PFGWing) machine is also slower than a modern machine, that will increase the optimum depth above that.

Once you have an optimum depth calculated, if you feel you don't have the resources to sieve the entire n=33.9K-100K range to the optimum depth, let me know and I'll assist with sieving. Since Ian has taken the Sierp side up to n=100K, we should at least sieve the Riesel side up to that depth even if we don't test that high right now.

Gary

Thanks for all the info, and no I didn't know that sieve file existed. Thanks, that'll speed things up a lot!
At n=80170, (70% of the way from 33.9K to 100K) PFGW took 11:35 (m:s) to PRP test it. At 555G, I'm expecting about one factor per 2:22 (m:s). So I've still got quite a ways to go. After I've started on it and have a rough idea of how fast it's going, I might ask for your help sieving. (at this point, I'm thinking I will)

Another thing I thought of. Typically like I suggested before, I wouldn't recommend breaking off a range for a 3X ratio of high-n to low-n in a sieve file. But in this case, I think we probably should. My new suggestion is to break off n=33.9K-60K. That will be a good piece for just you to test. Doing that means you won't have to sieve the entire thing so far. So plan of action:

1. PFGW a candidate at n=52.2K. (70% of 33.9K-60K range)

2. Sieve n=33.9K-100K until remove rate equals test time of above.

3. Break off n=33.9K-60K for testing.

4. When closing in on completion of testing, remove k's from n=60K-100K with primes and sieve that to the appropriate optimum depth using the timing of an n=88K (70% of n=60K-100K) candidate for the sieve removal rate. (Others including myself could do this.)

5. Test n=60K-100K.

In this case, I think this makes much more sense since your machines will have much different timing than mine.

If you can compute that optimum depth, I'll start helping with sieving. I'm thinking we won't need to sieve too much further using the lower PFGW timing. I may even sieve much above your optimum depth if it doesn't take too long. Technically I suppose it makes the most sense for me to sieve until MY removal rate is the same as YOUR PFGW time since you'll be doing the testing. I'll see what I think when I get into it.

Sieving large ranges is such an art! :smile:

Gary

Sounds like a good plan. It took me 4.08333... minutes to PFGW a candidate at n=52170, (for reference, 70% of n=33.9K-100K, n=80170, took 11.58333... minutes) which is almost exactly equal to my removal rate at 950G. I'm thinking I'll just sieve to 1T like I was planning and let you do any additional sieving you think would be good at the time, given the plan and my rates.

OK, I'm guessing I can sieve P=1T-2T, assuming that MY removal rate at P=2T will about equal YOUR n=52170 testing time. I'll experiment with it a little later tonight and let you know what I'll do.

How long for you to finish sieving to P=1T? That will dictate whether I put 1 or 2 cores on it or perhaps whether I sieve to P=1.5T or something else.

Here is what I get on one of my main Intel machines:

Removal rate at P=555G: 1 every 73.5 secs. (Using the much more accurate expected # of factors figure.)

Testing of an n=52170 candidate: 145 secs.

145 / 73.5 = 1.97 * 555G = 1095G

Therefore the optimum sieve depth using a relatively modern machine is P=~1.1T, quite close to your calculation.

I calculated the optimum if we used my sieving rate and your testing rate at P=1.85T, close to my previous guess of P=2T. I'm not quite willing to sieve that much. A P=1T range would take ~12 CPU days or P=850G range ~10 CPU days. I would be willing to sieve a P=500G range if you want, which would take ~6 CPU days.

Based on this, let me know how much and what range you'd like me to sieve. If you'll be done with your sieving in < 6 days and would still like me to do a full P=500G range, I'll put 2 cores on it to finish in 3 days.

If you give the word before then, I'll start on it mid-afternoon Sat.

Gary

I'm expecting to take 5x2 CPU days to get to 1T. I started last morning, then paused it for 24 hours to take part in a challenge for PSP sieving. So I'll expect it to finish in about 5 more days, which is Thursday the 19th. Since this is right on the edge of your < 6 days number, I'd only be waiting one day for your sieving to finish. I wouldn't really mind waiting another day if it's a lot more trouble to run it on two cores than just one.
:blahblah:
To summarize and clarify:
Go ahead and sieve 1T-1.5T, please. :smile: One or two cores, your choice.