Quote from: mikaelh on July 25, 2013, 11:15:28 PM

-hp8 is finally out!

Changes in -hp8:
* IMPORTANT: Fixed loss of potential blocks caused by fractional length calculation being skipped if fast divisibility test succeeds. (Thanks to mtrlt for spotting this.)
* Skip fractional length calculation for the first number in a chain (as suggested by gatra).
* Added a new configurable round primorial adjustment system.
* Round primorial can be adjusted through the -roundsievepercentage parameter (default value is 30, minimum is 1, maximum is 100). The parameter depends how much time is spent running the sieve. By default 30% of time is spent in the sieve and 70% is spent on checking the candidates produced by the sieve.
* Lots of other performance improvements.
* Bigger sieve sizes should no longer crash on Windows.
* Added new RPC commands setsievepercentage and setroundsievepercentage for changing parameters on the fly.
* GMP is now included as a separate library in the binary releases.

Links to downloads are on the first page. This release was delayed by a few bugs but it's finally here. Many improvements and fixes were included in this one. My own benchmarks show that block rate is nearly doubled compared to -hp7 on testnet. Performance on mainnet is also looking really good.

There's a new tuning parameter called -roundsievepercentage. You shouldn't need to touch this normally. Optimal value for mainnet seems to be around 30%. For testnet it's around 20% for some reason. The round primorial is adjusted based on this parameter. Smaller percentage means a bigger round primorial. Bigger round primorial in turn means that primes are more probable but the numbers will also get bigger which will slows down the testing.

Bigger sieve sizes seem to have suffered a performance hit with this release. I suggest starting with the default size.

Thanks mikaelh... Is there anything similar to Sunny's original gensieveroundlimitms as a fallback to limit the total time spent on a round?