To James Heinrich

Any news on the CPU efficiency calculator (http://mersenneforum.org/showthread.php?t=11784)?

Perhaps I should have been more specific. To give some pointers:

1) The suggestion given in http://mersenneforum.org/showpost.ph...67&postcount=8 really interested me, but I haven't heard progress in that area. Can it be implemented?

2) While the FFT 'timings' do apply to L-L, P-1 and ECM testing, it really seems (at least to me) that the GHz-day credit differs significantly from one worktype to another. In other words, the "FFT credit" would mainly apply to L-L and only very roughly to the other worktypes. Is it possible to discriminate the credit between them?

3) Some benchmarks sent to mersenne-aries, which are the basis for the calculation, have statistically improbable differences from the rest (see http://mersenne-aries.sili.net/bench...orderby=fft640, for example), and are probably affecting the results. Perhaps you could remove the doubtful ones.

[S485122]

Originally Posted by Unregistered

Some benchmarks sent to mersenne-aries, which are the basis for the calculation, have statistically improbable differences from the rest (see http://mersenne-aries.sili.net/bench...orderby=fft640, for example), and are probably affecting the results. Perhaps you could remove the doubtful ones.

Do not forget that for a same CPU at the same speed, memory speed plays a paramount role. This means that the benchmarks should be qualified by bus speed, memory multiplier, speed and latencies...

Jacob

[cheesehead]

Originally Posted by S485122

Do not forget that for a same CPU at the same speed, memory speed plays a paramount role. This means that the benchmarks should be qualified by bus speed, memory multiplier, speed and latencies...

... plus comparison aids such as (FFT time)/(bus speed).