P90 CPU years credits

[Carlo Monari]

I noticed that the credits for trial factoring are not homogeneus with the ones for P-1 factoring; it looks like trial factoring credits maybe fifty times more than P-1 factoring. Is that correct, and if yes, why?

I noticed also that in one case prime95 tried to start a LL test with a 1024k FFT, but then stopped because of rounding errors and restarted with a 1280k FFT (The exponent was 20256751). When the test ended, the credit was something more that two P90 CPU years, which means that the calculation was based on 1024k FFT timings, and not on 1280k FFT.
The test ran on a CPU with SSE2; a CPU without SSE2 would have used a 1024k FFT without problems.

[patrik]

I think that a P-1 always gives 0.001 years if you don't find a factor.

Also, AFAIK the LL credits are based on the default non-SSE2 FFT limits. The benchmark page shows these FFT crossover limits. They may be adjusted by the program after 1000 test iterations (see "New features in Version 22.8 of prime95.exe" in whatsnew.txt).

So the largest exponents currently handed out by the server lie in the range 24.93M-25.35M where SSE2 computers get less credit than what is motivated by the effort.

[PrimeCruncher]

Originally Posted by Carlo Monari

I noticed that the credits for trial factoring are not homogeneus with the ones for P-1 factoring; it looks like trial factoring credits maybe fifty times more than P-1 factoring. Is that correct, and if yes, why?

Nothing is really homogeneus with anything else. P-1 always gives 0.001 credit I believe, and TF gives about a tenth of the credit you would get from an LL test. This is to motivate people to do LL tests and because TF and P-1 aren't as useful; they can prove an exponent is composite, but not that it is prime.