At 11:20 AM 7/26/2003 +0100, Doug Feather wrote:
  I've currently got an exponent of the order of 20.05M running on a P4
(1.8Mhz).  It has decided to run it with FFT size 1280K.  It's done just
over 1M iterations (and hasn't finished Stage 2 P-1 since this only runs for
a few hours each night).  If I'd of know that it was going to use this FFT
size before it started, it would have been better for me to switch it to
another processor type - since then it would have used a smaller FFT size.
If I now move the saved file to another processor type would it use a
smaller FFT size or is the FFT size hard coded into the save file or is this
a bad idea to move the saved file for other reasons?

I think you can move it to another machine without problem. The FFT size is encoded in the save file, but when the save file is opened prime95 checks to see if a different FFT length should be used and converts if necessary.

The only downside to moving an exponent is that if *either* computer has
hardware problems, then your result could be corrupted.

You could also look in undoc.txt for options on making prime95 choose FFT
crossover points more aggressively. Look in results.txt for info on how close
prime95 was to choosing the 1024K FFT length. If it was a really close call
you could edit local.ini to make it choose the 1024k FFT length for that exponent.

While on the subject of FFT size where is the information stored about
which FFT size tests have been run on each machine?

This information is not recorded anywhere - my database does not care which FFT length was used on each LL result.

Unsubscribe & list info --
Mersenne Prime FAQ      --

Reply via email to