Today, we finished sending out new PPS Sieve workunits for 3M <= n < 6M. That's been sieved to p=400P. Meanwhile, manual sieving had been working on 6M <= n < 9M and had taken it to p=11.55P. As of a few minutes ago BOINC is now sending out work on n=6M-9M and manual sieving is working on 9M-12M.

What does this mean to the average user? Not much - it's not something you'd necessarily notice. It's possible that the workunits will take a bit longer in the beginning as more factors are being found, but for the same p range these tests take the same amount of time as the previous n range did. No credit adjustment is needed.

I've gotten several private questions about what this means. There is nothing wrong with the older workunits sieving the 3M-6M n range. The validator can tell which n range goes with which job and will always handle them properly. Those older workunits will continue to be sent out until they're all completed. There's no final deadline for when they have to be back.

Workunits might take a little longer than you're used to. But, when the 3M-6M was at this same p level they took longer too. We did not decrease credit as those workunits gradually took a shorter time so we're not increasing credit now. When benchmarking the new range vs the old range the difference was less than 2 seconds.

Sieving eliminates the need to run LLR tests by proving that some candidates are not prime. It does that by finding a factor for the candidate. Fewer factors are found as we search higher and higher. At some point it takes more time to eliminate a candidate by sieving than it takes to LLR test it. When that happens the sieve is said to be at its optimal depth. That's what happened with the 3M-6M range.

We'll be running the 6M-9M range for years, as those numbers are much larger and take longer to LLR test. You'll notice that instead of finding at most three factors per workunit you're now finding somewhere around 65. This will gradually drop off as the sieve goes higher over the next few years.

On my gtx570 on ubuntu 13.10 at stock clocks on the gpu, I note a change from about 1075 seconds per WU (wall clock) to roughly 1177 seconds. CPU time has dropped from about 68 seconds average per WU to about 37 seconds... more "oomph" for LLR available, which is good. All just due to the nature of the work, since there has been no software change for this.

No complaints here... just an observation. "Your mileage may vary", of course :-)

--Gary
____________
"I am he as you are he as you are me and we are all together"

87*2^3496188+1 is prime! (1052460 digits)
4 is not prime! (1 digit)