Help needed for doublecheck sieving!

[Mini-Geek]

When you say this is doublecheck sieving, do you mean that this range (1-1000 k, 100K-260K n) has already been LLRed, but not sieved? It seems that it would be rather inefficient to LLR this range if it was unsieved (not a word?)...or did it previously have both sieve and LLR done, and now we're repeating both processes?

[mdettweiler]

Originally Posted by Mini-Geek

When you say this is doublecheck sieving, do you mean that this range (1-1000 k, 100K-260K n) has already been LLRed, but not sieved? It seems that it would be rather inefficient to LLR this range if it was unsieved (not a word?)...or did it previously have both sieve and LLR done, and now we're repeating both processes?

The entire range was both sieved and LLRed previously. However, that was all done by various individual k reservations, rather than through an organized project like NPLB, so we don't have a "big sieve file" to go back to and re-LLR when we want to doublecheck. Thus, we have to sieve this range again before we can doublecheck it with LLR.

Once we've finished this doublechecking to n=260K, however, we'll have access to NPLB sieve files for 300<k<1001 for future doublechecking efforts (since NPLB's sieve files start at n=260K), as well as first-pass residuals to compare our doublechecks with--which will eliminate the need to have exclusively known-stable machines to be doing the doublechecking, since we won't have to be completely reliant on our doublecheck results.

[gd_barnes]

Originally Posted by Anonymous

Don't worry, stable overclocked machines are still welcome. The only requirement is that you run a Prime95/mprime stress test for a few hours to see whether your machine will produce good residuals (it probably will).

Please note, however, that even machines that produce bad LLR residuals will still do just fine with sieving.

Yes, overclocked machines are no problem in sieving. Even if they miss a factor or two, that would only cause us to spend slightly longer LLRing but we wouldn't miss any primes. I've never heard any reports of 'false factors' even from problematic machines.


Gary

[mdettweiler]

Taking 300G-310G.

[em99010pepe]

Originally Posted by gd_barnes

Yes, overclocked machines are no problem in sieving. Even if they miss a factor or two, that would only cause us to spend slightly longer LLRing but we wouldn't miss any primes. I've never heard any reports of 'false factors' even from problematic machines.


Gary

With my experience sieving reduces in 5 ºC the temperature of the cores. I can easily sieve at 3.1 GHz with my quad-core but not LLRing, the machine just reboots. It's stable LLRing at 2.9 GHz...

[mdettweiler]

Originally Posted by em99010pepe

With my experience sieving reduces in 5 ºC the temperature of the cores. I can easily sieve at 3.1 GHz with my quad-core but not LLRing, the machine just reboots. It's stable LLRing at 2.9 GHz...

Not only is the temperature a little less when sieving (because the FPU is not as heavily used), sieving is also less sensitive to slight errors than LLR.

[jasong]

310G-910G reserved

(If wblipp is reading this, my ecm numbers are still reserved, but you can tell me to abandon them if you get impatient. :) )

[mdettweiler]

Originally Posted by jasong

310G-910G reserved

(If wblipp is reading this, my ecm numbers are still reserved, but you can tell me to abandon them if you get impatient. :) )

Wow, that's a big range! Please keep in mind that this sieve has 500 k's in it, though, so a given range will take quite a while compared to a similar-sized range on a "normal" sized sieve. But, if you can handle it, that's awesome!

[gd_barnes]

Originally Posted by jasong

310G-910G reserved

(If wblipp is reading this, my ecm numbers are still reserved, but you can tell me to abandon them if you get impatient. :) )

Jasong, thanks for the help. I need to give you some particulars so that you're aware of what you're reserving:

Previously with 351 k's, Chris (Flatlander) was getting 230K P/sec on a high-speed machine. With 500 k's I'm estimating the P-rate on a high-speed machine now to be 20% less or 184K P/sec.

Assuming you're running a high-speed machine, at 184K P/sec, P=600G would take you 3.26M CPU secs. or 37.7 CPU days. So if you put 4 high-speed cores on it running 24x7, it would still take you ~9.5 calendar days.

Are you willing to commit to 4 cores for 9.5 days for sieving (or 2 for 19 days) running 24x7? If so, that's great and we appreciate it very much. I just want you to be aware of the time involved.


Thanks,
Gary

[gd_barnes]

Anon,

What is your P-rate per second at P=300G on this sieve?


Thanks,
Gary

[Flatlander]

For information, my rates are (per core):
160,000 p/sec
9 sec per factor
C2D 3000MHz
(About 4% better when one core is idle.)