Prime95 30.8 (beta, big P-1 changes, see post #551)

[axn]

Originally Posted by ATH

Yeah I know, but I thought I read that the formula was improved a bit in the most recent builds of 30.8 ?

Credit calculation is done server-side. Nothing can be done on the client side. You're perhaps confusing it with improved "cost estimation" which is needed to select optimal P-1 bounds.
Server logic needs to be fixed.

[ATH]

Originally Posted by axn

Credit calculation is done server-side. Nothing can be done on the client side. You're perhaps confusing it with improved "cost estimation" which is needed to select optimal P-1 bounds.
Server logic needs to be fixed.

Thanks, yes it was "cost function" I mistook for the credit calculation.

[petrw1]

When there are more Total Workers than MaxHighMemWorkers (i.e. Stage) ...
and if Stage2 falls behind v30.8 alternates workers to keep the backlog on each relatively equal.

The previous version would completely catch up 1 worker while they others madly race on then move on to another to catch up ... etc.

[tuckerkao]

Originally Posted by axn

Credit calculation is done server-side.

The P-1 with the same B1 and B2 of the same exponent will yield around 15% more GHz day credits from 30.8 v14 compare to 30.7 v9.

[axn]

Originally Posted by tuckerkao

The P-1 with the same B1 and B2 of the same exponent will yield around 15% more GHz day credits from 30.8 v14 compare to 30.7 v9.

I am not sure what the relevance of this observation is to the quoted part (about credit calculation being done server-side).

[tuckerkao]

Originally Posted by axn

I am not sure what the relevance of this observation is to the quoted part (about credit calculation being done server-side).

As soon as Prime95 submitted the P-1 results, the server returned the total amount of GHz day credits I earned to my software client end, so the server must be the one that decided the credit amount based on the different Prime95 versions.

I had 1 machine with Prime95 30.7 v9 and the server always rewarded around 30 GHz days per NF-PM1, the other machine with Prime95 30.8 v14 and the server always rewarded around 35 GHz days per NF-PM1. If I unplugged the internet router, then nothing would happen.

Just run the P-1 of any small exponent, see the difference of GHz days credits given from the server to Prime95 30.7 v9 and 30.8 v14, take only several minutes. The only problem is that the difference may not be 15% because it's a different range.

The most relevant P-1 records will have to show all the P-1 exponents which I've finished during the past 2 weeks. UncWilly made the rule(SRBase Thread) which I must post that on my own thread. Thus I'm not going to post that screenshot here until UncWilly says is okay to do so.

[tuckerkao]

I'm wondering when Prime95 30.8 v15 will be released and whether the speed of PRP tests for the larger exponents will be improved or not?

[axn]

Originally Posted by tuckerkao

I had 1 machine with Prime95 30.7 v9 and the server always rewarded around 30 GHz days per NF-PM1, the other machine with Prime95 30.8 v14 and the server always rewarded around 35 GHz days per NF-PM1. If I unplugged the internet router, then nothing would happen.

Just run the P-1 of any small exponent, see the difference of GHz days credits given from the server to Prime95 30.7 v9 and 30.8 v14, take only several minutes. The only problem is that the difference may not be 15% because it's a different range.

30.8 uses a bigger FFT for stage 2. This information is included in the submitted result, so the server grants bigger credit. Obviously the difference in credit is dependent on the difference in FFT (vs normal FFT used by 30.7) - this is not a fixed %

Originally Posted by tuckerkao

I'm wondering when Prime95 30.8 v15 will be released and whether the speed of PRP tests for the larger exponents will be improved or not?

AFAIK, there are no planned speed improvements for PRP

[Zhangrc]

Originally Posted by tuckerkao

The P-1 with the same B1 and B2 of the same exponent will yield around 15% more GHz day credits from 30.8 v14 compare to 30.7 v9.

30.8 uses a larger FFT in stage 2.

Originally Posted by tuckerkao

I'm wondering when Prime95 30.8 v15 will be released and whether the speed of PRP tests for the larger exponents will be improved or not?

That's George's decision. He is so talented, don't put extra pressure on him and he'll give us a surprise!
Maybe it will be 30.9b1, because George said that he is planning to implement the new stage 2 algorithm in ECM.

[James Heinrich]

I have some other important work running so Prime95 (30.8b13) has been suspended in from PauseWhileRunning for many days now. However at least twice I've happened to notice that it had silently crashed (or, at least, it was no longer running). I didn't see any crash messages or anything though.

[storm5510]

Originally Posted by tuckerkao

The P-1 with the same B1 and B2 of the same exponent will yield around 15% more GHz day credits from 30.8 v14 compare to 30.7 v9.

I have ran the same bounds on ECM's and then use GMP-ECM to run Stage 2. I don't see the value is doing this with P-1...