Um, 28.5 just came out. Great for Haswell CPUs.
Type: Posts; User: prime95
Um, 28.5 just came out. Great for Haswell CPUs.
That 0.4375 is just the kind of round off error I wouldn't worry about. BTW, what number are you testing.
Again look at the size of the roundoff errors. 0.40625 and 0.4375 are OK, 0.4999 is really bad.
I think a roundoff error of 0.40625 and 0.4375 may sometimes not be reproducible because there...
Can you post the numbers that are getting roundoff errors? I'll look into whether the FFT size selection code needs a little fine tuning.
If you are getting small reproducible roundoff error values like 0.40625 and 0.4375 then it is likely due to testing a value near the limits of what that FFT size can test. In other words, nothing...
This is probably normal. On a Pentium 4, some FFT lengths are actually faster using FFTs optimized for the Core 2. Prime95 knows this and selects the right one accordingly. Admittedly, the "Core2...
Yes. A quick edit of worktodo.txt will fix the problem
Your best configuration is likely to be 4 worker threads with no multi-threading. Use Test/Worker-windows to change turn off the multi-threading.
Time it to see which is faster.
Prime95 version 26.2 is available for beta testing. See http://www.mersenneforum.org/showthread.php?t=13887
It wouldn't hurt to have an SoB user or two give it a try.
Has anyone else noticed this problem? I just tried replicating the problem and couldn't. Please check your results.txt files and let me know if this was a one time occurrence or if finding P-1...
If you have 128MB of physical memory, then you shouldn't tell the program to use more than 96MB or so (assumes OS and various programs need 32MB). If you tell the program to use too much memory, you...
The optimal bounds picker changed too.
I remembered what the P4 optimal bounds bug was - the routine addr_offset in gwnum.c used uninitialized global variable FFTLEN rather than the local...
Two observations:
Sbfactorer does not seem to be based on the v23.4 gimps sources. Stage 2 was rewritten two months ago to use more memory if it can, use fewer multiplies, and to be more...
Official verification is always done using different hardware and software. In this case a 1 GHz MIPS machine using Ernst Mayer's Mlucas program
GIMPS is currently testing exponents near 20,000,000 - so you can be sure the new find is under 6 million digits.