Sieving discussion thread

[biwema]

Originally Posted by jasong

Question: If you set it to log factors, will it log factors as soon as it gets them, or go by the update interval? The reason I ask is that I set the update interval to 10,000 minutes, thinking that only applied to the main file. I'm hoping the factors will update more frequently, and if there's a power outage I can simply update the starting point with vi.

Is this the correct way to do things?

Yes. The factors are written immediately.
If you set the update interval to a high level, you risk to lose the whole sieving time after the last factor was found. In this sieving project the factor density is so high that it will be no problem, at all.
Besides that you save the intermediate file saving time (around 90 sec every time)

[Skligmund]

I'm averaging about 3.5-4.0 k/sec (times 2) sieving 333,333 on my X2 3800+, how is everybody else doing? I just want to benchmark myself is all.

[gribozavr]

Celeron 1200Mhz: 14.5-15.5 k/sec
Athlon 2500+ 1800Mhz: 4-6 k/sec
Athlon64 3000+ 1800Mhz: 4-5.5 k/sec

[jmblazek]

I must be doing something wrong. My AMD Athlon(tm) 64 3400+ is only doing 1 k per 2.5-3.0 seconds. Any ideas???

I'm currently working on 500T-550T.

[Skligmund]

I mis-typed..

3.5-4 sec/k

[R. Gerbicz]

Originally Posted by jmblazek

I must be doing something wrong. My AMD Athlon(tm) 64 3400+ is only doing 1 k per 2.5-3.0 seconds. Any ideas???

I'm currently working on 500T-550T.

Sieving k/sec rate depends on the sieving depth, you can sieve about 2 times slower at 500T than if you would sieve at 250T.

[jmblazek]

Originally Posted by Skligmund

I mis-typed..

3.5-4 sec/k

Whew...that makes sense now...since you're working about 300T deeper.

[Cybertronic]

Originally Posted by MooooMoo

The problems with both ideas is the huge size of the new output file and the factor list.

Is not the problem, I think. We can write a program that give us all reduce numbers ( The number is not in the new file after sieving ) The MASTER ADMIN taks the small file and reduce from file 32ye9b also with a special program. I work with 2 PCs and this is for my project very helpfull when I sieve serveral ranges.

best

[MooMoo2]

For n=195000, we actually sieved far more than the previous record holders did. They sieved only to 281T, while we got over 1000T (see http://www.hpc-europa.org/index.php?...i_high/csajbok)

The number of Intel Itanium 2 processors they used "varied between 1 and 96" while they were working on getting their previous record twin.

They had a total of 5 million candidates after sieving (out of an original 8.5 billion). This means they were probably choosing a range that had an ~80% certainty of containing a twin. It would be equivalent to sieving n=195000 from 1-22G.

[Skligmund]

Originally Posted by MooooMoo

- somewhere to upload the new output file, so that NewPGen's feature for " merging files across several machines" can be used.

I can't figure out how to use that on my own computers. The instructions are confusing at best. I know this is the wrong thread for this, but it only made sense here. If anybody knows how this works, let me know. Maybe make a simple tutorial for it?

[gribozavr]

If you want to report results, you don't need to upload the datafile; only newpgen.del file is needed. Of cource, unless you forgot to set newpgen to log factors.