New credit parameters for some exponents

[S485122]

It is only normal that the credit awarded takes into account the new FFT sizes introduced by version 26.3.

I have a problem though : I recently completed a doublecheck on M25883279, according to the pre 26.3 FFT sizes and credits I should have received 25,7395 GHz day credits. PrimeNet only gave 22,8205. So instead of 0,08592 GHz seconds of credit per iteration PrimeNet used 0,076176253 GHz seconds (because of rounding it is something between 0,076176086 and 0,076176420.) In the sources of Prime95 26,3 I cannot find that value.

Is it possible to publish the new credit computation tables ?

Jacob

[garo]

Prime95 reports the FFT size that was used and Primenet gives credit accordingly.

[Kevin]

I have a question along the same lines. Roughly since upgrading to version 26.3 from 26.2, I've been getting about .5GHZ days less on P-1 tests in the same exponent range with the same B1/B2 settings. I understand that Prime95 using different FFT sizes will lead to different amount of credits, but that change should have happened when I upgraded to 26.2, not to 26.3 . I can't think of any good reason for why the change in credit has only occurred recently.

[Prime95]

Until a few days ago, Primenet looked up the FFT length in a timings table. This table had timings for all FFT lengths that are 5,6,7,or 8 * a power of 2 (that is all the FFT lengths that version 25 supported).

Version 26 was released with new FFT lengths. What happened when you reported a result with a FFT length not in the table? The server used an alternate method to get the timing by looking at the exponent that was tested.

An example: exponent = 52M, v25 FFTlen = 3M, v26 FFTlen = 2800K. Prime95 v25 uses 3M FFT and gets credit for a 3M FFT. Prime95 v26 uses a 2800K FFT, the server has no entry for that, the server thinks 52M exponents are tested with 3M FFTs, you get credit for a 3M FFT.

Earlier this week this changed. If a timing lookup fails, the server will interpolate the timings for 2560K and 3M FFTs to come up with a reasonable estimate for the 2800K FFT.

[petrw1]

My "gains" and "losses" probably balance.

I've had a few exponents start an exponent under 25.9 and finish quicker under 26.2 but get full credit based on the old FFT
AND
more recently some that got the lower credit of the smaller FFT even though some of the work was slower with the larger FFT and 25.9 code.

My P-1 core when first upgraded to 26.2 and the new FFT sizes got the same credit with about a 5% speed improvement.
AND
more recently the credit for P-1 in the same 53M range, same B1/B2 has dropped fom 4.4061 per to 4.1073: about 7% less.