Sieving for CRUS

[MisterBitcoin]

Since R3 sieve is done, i would start with sieving existing sieve files deeper for n=2,5K-10K. If i havent counted wrong those are 66 bases for the Riesel side alone. (and 49 on Sierp-side)

[gd_barnes]

Originally Posted by MisterBitcoin

Since R3 sieve is done, i would start with sieving existing sieve files deeper for n=2,5K-10K. If i havent counted wrong those are 66 bases for the Riesel side alone. (and 49 on Sierp-side)

Optimum sieve depth on those averages P=~15G and they are currently only at P=1G. That is a serious effort for most folks, even for just sieving. Yoyo cannot do them because it's too many k's.

You definitely want to only take one at a time. :-)

[MisterBitcoin]

Originally Posted by gd_barnes

Optimum sieve depth on those averages P=~15G and they are currently only at P=1G. That is a serious effort for most folks, even for just sieving. Yoyo cannot do them because it's too many k's.

You definitely want to only take one at a time. :-)

I just started with R1012, my speed is at 270p/sec using 4 threads. So yeah, its slow.

However in a few days my new 10-Core i9 will arrive, so i should get some speed-up^^

[MisterBitcoin]

Gary, what is your opinion on spliting larger bases, lets say more than 10K k´s, into smaller files. Example:

2,5K up to 4K
4K up to 7,5K
7,5K up to 10K.



That way should make it a bit easier to sieve deeper for the 2nd and 3th stage due to the removed k´s.
(Granted, thats more effort from your, Reb´s and my side)

[rebirther]

Originally Posted by MisterBitcoin

Gary, what is your opinion on spliting larger bases, lets say more than 10K k´s, into smaller files. Example:

2,5K up to 4K
4K up to 7,5K
7,5K up to 10K.



That way should make it a bit easier to sieve deeper for the 2nd and 3th stage due to the removed k´s.
(Granted, thats more effort from your, Reb´s and my side)

I will be using 10k factors per core next time but running 30 instances on 16core.

[gd_barnes]

Originally Posted by MisterBitcoin

Gary, what is your opinion on spliting larger bases, lets say more than 10K k´s, into smaller files. Example:

2,5K up to 4K
4K up to 7,5K
7,5K up to 10K.



That way should make it a bit easier to sieve deeper for the 2nd and 3th stage due to the removed k´s.
(Granted, thats more effort from your, Reb´s and my side)

If I was going to split n=2.5K-10K by n-value, if would be in two: n=2.5K-5K and n=5K-10K. Splitting more than that saves so little CPU effort that it is not worth the manual effort to do so.

But...the tests are so fast for n=2.5K-10K that I just prefer to split them up by k-value and sieve the entire n-range to the same depth.

For instance I sieved the 320,000+ k's for R271 by splitting it up into 20 instances of ~16,000 k's each and sieving all of n=2.5K-10K to P=1G. That is not nearly deep enough for testing but the files on the pages are intended to be a starting point for others to sieve deeper before beginning testing. Optimum depth for the typical base for the entire k-range and n=2.5K-10K is P=~10G to 15G. If I am personally testing them I'll just sieve the whole thing to P=15G.

[rogue]

Originally Posted by gd_barnes

If I was going to split n=2.5K-10K by n-value, if would be in two: n=2.5K-5K and n=5K-10K. Splitting more than that saves so little CPU effort that it is not worth the manual effort to do so.

But...the tests are so fast for n=2.5K-10K that I just prefer to split them up by k-value and sieve the entire n-range to the same depth.

For instance I sieved the 320,000+ k's for R271 by splitting it up into 20 instances of ~16,000 k's each and sieving all of n=2.5K-10K to P=1G. That is not nearly deep enough for testing but the files on the pages are intended to be a starting point for others to sieve deeper before beginning testing. Optimum depth for the typical base for the entire k-range and n=2.5K-10K is P=~10G to 15G. If I am personally testing them I'll just sieve the whole thing to P=15G.

I would be curious to see some statistics comparing the speed of srsieve to srsieve2 for these ranges. How you split n or k is up to you for that comparison.

[MisterBitcoin]

Ok i have an short update about R1012. I started the process early this morning on one of my servers, after around 15 hours i found ~600.000 factors.
Earlier it showed to reach 5e9 on the 28.01, now it will reach this in about 24 hours.


I think its safe to say its ready for n=2,5K-5K by tomorrow.

Also i finished R1005, at around 12e9 it surpassed the average LLR runtime by ~20%. Should be enough to get tested by BOINC. Should i send the file to reb or Gary?

[gd_barnes]

Originally Posted by MisterBitcoin

Ok i have an short update about R1012. I started the process early this morning on one of my servers, after around 15 hours i found ~600.000 factors.
Earlier it showed to reach 5e9 on the 28.01, now it will reach this in about 24 hours.


I think its safe to say its ready for n=2,5K-5K by tomorrow.

Also i finished R1005, at around 12e9 it surpassed the average LLR runtime by ~20%. Should be enough to get tested by BOINC. Should i send the file to reb or Gary?

Please wait to send me R1012 until you are done with the entire n-range. There is no reason to send me pieces. For such small tests it would be easier if the entire n-range was sieved to the same depth even if it means a little more CPU time in the long run. It's easier to set it and forget it until it is done, both for sieving and testing.

For R1005, BOINC does not run such small tests. You can send me the sieve file. I will post it on the page. Others can reserve it if they want although I doubt it will get much interest for such a large base and large # of k's remaining. Anything with over 10,000 k's remaining is an extremely large effort to fully test. Even over 5,000 k's is big.

[MisterBitcoin]

Originally Posted by gd_barnes

Please wait to send me R1012 until you are done with the entire n-range. There is no reason to send me pieces. For such small tests it would be easier if the entire n-range was sieved to the same depth even if it means a little more CPU time in the long run. It's easier to set it and forget it until it is done, both for sieving and testing.

For R1005, BOINC does not run such small tests. You can send me the sieve file. I will post it on the page. Others can reserve it if they want although I doubt it will get much interest for such a large base and large # of k's remaining. Anything with over 10,000 k's remaining is an extremely large effort to fully test. Even over 5,000 k's is big.

Sieving R1012 up to P=15e9 shouldn´t take much longer, expecting it to finish in about 6-7 days before taking the next file.

R1005 sieve file will be added in the thread shortly.

[MisterBitcoin]

R1012 reached p=15e9 at a speed of 5,77 secs/factor. The sieve file has been send via PM to Gary.



Next file taken for sieving is R375.