[quote]Yep! That's the one. Is there any way to reduce that 150 wasted minutes?[/quote]
As noted above by Prime95 (George), you can reduce the time between save file writes from 30 minutes (default) to a shorter time. That would reduce wasted time retesting by reducing the average time lost per error from an average of 15 minutes (half the disk write interval) to a shorter average amount (= half of the new, shorter disk write interval). However, that change would simultaneously increase the total amount of time wasted for performing the larger number of save file writes.

But note that the number of roundoff > 0.40 errors per run, 10, in the given example, and thus the 10 x 15 = 150 minute time loss is only an example, not a hard-and-fast figure. Fewer such errors per run means less time waste. However, just as in the preceding paragraph, any measure you take to reduce that number might simultaneously increase the time used/wasted for some other reason, so it's important to look at the total picture.

150 minutes is < 1% of an 11-day run.

Here is a question about Prime 95 versions for George. Last December, I won an AMD Athlon 2400+ Throughbred version B at one of AMD's performance events. My MB would not support this chip until the latest BIOS version was released (Mar 03). When I finally got this chip to work in my MB, I initially could not get it to run at full speed (i.e. 2000MHz). Consequently, I never bothered to upgrade the speed set in Prime 95 version 21.4 from the 1466MHz that it was set to for my AMD 1700+ Palomino.
Iteration times for M18974113 did not change very much regardless of what speed or FSB my computer was at. They ranged from about .126 ~ .140 regardless of whether my computer was running at 1500MHz (15x100), 1700MHz (16.5x103), or 2000MHz (15x133). Yesterday, when I finally figured out how to run at 2000MHz (15x133), my iteration times were .126. Then I upgraded to Prime version 23.3.1 and my iteration times dropped to .092 ~ .094. Any ideas? More importantly, can other Athlon owners get a similar performance boost by upgrading Prime 95 to a newer version like I apparently did?

I think the problem was in the older versions timing code. In v21 the CPU clock frequency was set manually and this was used to give you iteration times. So I think the iteration timings were wrong. If you have any odl logs you may want to see the actual amount of time sent doing 1000 iterations and you will probably find that it did vary as expected with clock speed.

I may be wrong but I vaguely remember this being brought up before.

[quote="garo"]I think the problem was in the older versions timing code. In v21 the CPU clock frequency was set manually and this was used to give you iteration times. So I think the iteration timings were wrong. If you have any odl logs you may want to see the actual amount of time sent doing 1000 iterations and you will probably find that it did vary as expected with clock speed.

I may be wrong but I vaguely remember this being brought up before.[/quote]

Yes, you are correct, see [url]http://www.mersenneforum.org/viewtopic.php?p=649[/url].

Work in Progress...
 

If I upgrade, will the new client pickup the work in progress or start over? :?

Yes.

Shut down Prime95, then unzip the new client into the Prime95 folder. Then start it up.

Voila!! Updated!!

can we clean out some of the old stuff in here and make it just the latest client

have only the current version

( mabey the last 2 or 3 versions just incase a bug is found)

Good idea...

George-

Please post new announcements here and in the sticky thread... Delete the old announcement in the sticky thread if you want... That thread is locked so you will have to unlock it first, edit/add your changes, and relock it...

That particular thread # is linked all over the place so I didn't want to kill it and start a new one...

Let me know if this works for you...

Re: Work in Progress...
 

[quote="forte"]If I upgrade, will the new client pickup the work in progress or start over? ?[/quote]

If you upgrade to a different client version number, IME the client will restart - it doesn't seem to like changing version numbers in midstream.

If you upgrade to a different OS and run THE SAME version number, it will pickup the work in progress.

That's why I've not gotten around to upgrading my P4 past 23.2 yet - it's pretty far along on a 10,000,000+ digit LL....

I?ve never had a problem just stopping, overwriting the executable, and restarting in the middle of exponents.