Help needed for doublecheck sieving!

[mdettweiler]

Originally Posted by AES

Is there a reason you're not using sr2sieve?

I'll take 30e9 to 50e9

srsieve is generally faster when you have more than 60-70 k's in the sieve.

Thanks for helping out!

[AES]

You may want test it on this sieve. I have similar hardware and with sr2...

64-bit will do 10e9 in less than 10 hours

32-bit will do 10e9 in less than 16 hours

[mdettweiler]

Originally Posted by AES

You may want test it on this sieve. I have similar hardware and with sr2...

64-bit will do 10e9 in less than 10 hours

32-bit will do 10e9 in less than 16 hours

Hmm. Okay, I'll check it out.

In the meantime, feel free to use either srsieve or sr2sieve, whichever one you find faster--they both produce compatible results, so it won't make any difference on my end.

[gd_barnes]

It is possible that sr2sieve is now faster with 351 k's like this. It may also depend on the type of machine your using. When I started the huge sieve for 400<k<=1001 (301 k's) for n=260K-600K, srsieve was faster but that was before the most recent version of sr2sieve that Geoff put out there on 12/14/07.

The newer sr2sieve for the 50 k's for 300<k<400 for n=260K-1M is clearly faster.

So yeah, try both on your machine for these 351 k's.

BTW, the estimated optimum of 550G was an 'extraploted' guess. It assumed that an LLR test of a candidate at n=212K (70% of the n-range of n=100K-260K) took 60 secs. and it took into account the removal rate that Anon was getting on his machine at P=~19G. I didn't actually run the test. Anon or whomever, as we get past P=~200G, we'll see if we can nail down a better estimated optimum sieve depth. Technically with n-max being > 2x n-min, it's a little more efficient to break off a large chunk while continuing to sieve the higher n-ranges but personally, I don't think it's worth messing with at these lower n-ranges.

In the mean time, Anon, can you run an LLR test on your machine for a candidate around n=212K? That'll give a removal rate to shoot for on a comparable machine.

After 300<k<400 is done sieving to P=6T (and breaking off n=260K-600K) towards the end of February, I'll put 3 cores on this effort if it's not already done. That'll leave me 5-6 cores to continue 300<k< 400 for n=600K-1M for anticipated completion sometime in late March. At that point, we'll have 3 different n-ranges that people can test at for top-5000 primes. I'm hoping the higher n-ranges will draw in even more people who like to find much larger primes.


Gary

[Flatlander]

To use sr2sieve do I just download it and change "srsieve" in the .bat file?

[mdettweiler]

Originally Posted by Flatlander

To use sr2sieve do I just download it and change "srsieve" in the .bat file?

No, its command line switches are a little different. Your .bat file will need to look like this:

sr2sieve -p 10e9 -P 20e9 -i nplb-doublecheck-sieve_10G.txt

Other than that, it's pretty much the same.

[mdettweiler]

Originally Posted by gd_barnes

It is possible that sr2sieve is now faster with 351 k's like this. It may also depend on the type of machine your using. When I started the huge sieve for 400<k<=1001 (301 k's) for n=260K-600K, srsieve was faster but that was before the most recent version of sr2sieve that Geoff put out there on 12/14/07.

The newer sr2sieve for the 50 k's for 300<k<400 for n=260K-1M is clearly faster.

So yeah, try both on your machine for these 351 k's.

I'll try that sometime today.

Originally Posted by gd_barnes

BTW, the estimated optimum of 550G was an 'extraploted' guess. It assumed that an LLR test of a candidate at n=212K (70% of the n-range of n=100K-260K) took 60 secs. and it took into account the removal rate that Anon was getting on his machine at P=~19G. I didn't actually run the test. Anon or whomever, as we get past P=~200G, we'll see if we can nail down a better estimated optimum sieve depth. Technically with n-max being > 2x n-min, it's a little more efficient to break off a large chunk while continuing to sieve the higher n-ranges but personally, I don't think it's worth messing with at these lower n-ranges.

In the mean time, Anon, can you run an LLR test on your machine for a candidate around n=212K? That'll give a removal rate to shoot for on a comparable machine.

Okay, I'll do that soon.

Originally Posted by gd_barnes

After 300<k<400 is done sieving to P=6T (and breaking off n=260K-600K) towards the end of February, I'll put 3 cores on this effort if it's not already done. That'll leave me 5-6 cores to continue 300<k< 400 for n=600K-1M for anticipated completion sometime in late March. At that point, we'll have 3 different n-ranges that people can test at for top-5000 primes. I'm hoping the higher n-ranges will draw in even more people who like to find much larger primes.


Gary

Okay, cool.

Anon

[Flatlander]

20G-30G finished.

Taking 50G-70G.

srsieve 130,000 p/sec (20G-30G)
sr2sieve 230,000 p/sec (50G-70G)

[mdettweiler]

Originally Posted by Flatlander

20G-30G finished.

Taking 50G-70G.

srsieve 130,000 p/sec (20G-30G)
sr2sieve 230,000 p/sec (50G-70G)

Oh, wow!

I'll change the instructions in the first post so that they tell people to use sr2sieve. I'll also update the sieve file archive to include the updated batch file and shell script.

[mdettweiler]

Okay, I updated the first post in this thread to reflect the change to sr2sieve. The instructions are significantly different, so you'll want to read them again.

[Flatlander]

I'm already running this:

sr2sieve -p 50e9 -P 70e9 -i nplb-doublecheck-sieve_10G.txt

is that ok?