Prime95 version 26.2

[Mini-Geek]

Originally Posted by lpmurray

I can't believe people are so petty about increasing the score that the would steal other peoples work.

I don't know if people actually do that, but it's best to guard against it, in any case. Score aside, we wouldn't want to miss a prime, or otherwise have garbage data, (if not total garbage, at least not up to the mathematician's standard of close-as-possible-to-perfection) just because it was easy for some "doublecheckers" to do no work and just report the last person's residues.

[Rhyled]

I just completed 4 LL-D checks using 26.2 64-bit. All four successfully completed the double checks by verifying the previous LL run, so I'm happy with the accuracy of 26.2 for LL tests.
(M24309653, M24350327, M24441799, M24309661)

I'm also happy with the performance. No worse than a 0.25 round off error (that I saw). Approximately a 15% increase in calculation speed, less a 2% drop because I had SUM(Inputs) and Roundoff error checking marked. Per iteration time was ~ 16ms (i7-920 @ 3.71 GHz) for a 1280K FFT.

I'm trying ECM & ECM-F runs next - but unless I find a factor, that's going to be a less certain test than LL-D.

Nice job, George.

[JuanTutors]

I once requested that on multi-core computers, the LL test of a 100M digit Mersenne be started during the part of the P-1 factoring that does not take up all cores of the processor. (My argument was based on computer lifetime and human lifetime arguments).

Can you add this feature as an option that can be written into prime.txt?

[Rhyled]

I got lucky and found an ECM factor in my first dozen attempts on 26.2. (M5308217) I have no idea if the ECM speed is any different than the previous version, but the logic looks solid.

Back to my search for the 48th

[Mini-Geek]

Originally Posted by Rhyled

I have no idea if the ECM speed is any different than the previous version, but the logic looks solid.

It should be faster by about the same degree that LL tests are, since they all (LL/DC, P-1, ECM) use the FFT. The TF code wasn't improved, but the FFT code was.

[cheesehead]

Originally Posted by dominicanpapi82

I once requested that on multi-core computers, the LL test of a 100M digit Mersenne be started during the part of the P-1 factoring that does not take up all cores of the processor.

As Mini-Geek has just pointed out in response to another post, the main loop of P-1 performs FFT operations, just as the main loop of LL/DC (and ECM) does. So there's no intrinsic difference in processor core usage between them. They're all using the same FFT calculation subroutines. If P-1 is taking "up all cores of the processor", then it will do so practically all the time.

If you're referring to the non-FFT operations in P-1: they occupy only a very small slice of time, principally for the GCD calculation, which takes only a few minutes at most, at the end of stage 1 or stage 2.

Were you perhaps thinking of stage 2's large memory usage? That bears no relation to the number of processor cores used.

[diep]

Originally Posted by cheesehead

As Mini-Geek has just pointed out in response to another post, the main loop of P-1 performs FFT operations, just as the main loop of LL/DC (and ECM) does. So there's no intrinsic difference in processor core usage between them. They're all using the same FFT calculation subroutines. If P-1 is taking "up all cores of the processor", then it will do so practically all the time.

If you're referring to the non-FFT operations in P-1: they occupy only a very small slice of time, principally for the GCD calculation, which takes only a few minutes at most, at the end of stage 1 or stage 2.

Were you perhaps thinking of stage 2's large memory usage? That bears no relation to the number of processor cores used.

LL / LLR is just squaring in most cases and taking easy modulo.

So it's a simpler form of a FFT in case of LL/LLR when compared with most generic codes which need regurarly multiplication and not so easy modulo.

[hgj9999]

Does 26.2 support Windows 7? If so, both 32bit and 64bit are support?

[petrw1]

Originally Posted by Prime95

Known bugs and fixes:

... Fixed in 26.3 ...

Is there an estimated date for 26.3 to be available?

[Svenie25]

Originally Posted by hgj9999

Does 26.2 support Windows 7? If so, both 32bit and 64bit are support?

I myself use it under W7 64bit without any problems.

[cheesehead]

Originally Posted by diep

LL / LLR is just squaring in most cases and taking easy modulo.

So it's a simpler form of a FFT in case of LL/LLR when compared with most generic codes which need regurarly multiplication and not so easy modulo.

How is your comment related to processor core use during the P-1 main FFT loop in comparison to processor core use during the LL main FFT loop?