Weird TF Behavior Prime95 64 bit v24.13, WinXP 64
 

I have an AMD Athlon X2 6400 PC running at 3.2 GHz doing TF. This PC has been running Prime95 24.13 with few problems since Dec. 2007. Recently, I have been getting Prime95 application errors when Prime95 tries to factor a number greater than M61853387. The error occurs when Prime95 tries to write out the first save file after it starts factoring. I have sent the Error reports to Microsoft but have not received any feedback. No information is saved in the Results.txt or Prime.log files. Is there a start up parameter that I can use to capture more information?

Have you tried another version of Prime95? ([url=ftp://mersenne.org/gimps/p64v259.zip]v25.9.2[/url] is the most current x64 version right now)

[quote=James Heinrich;164655]Have you tried another version of Prime95? ([URL="ftp://mersenne.org/gimps/p64v259.zip"]v25.9.2[/URL] is the most current x64 version right now)[/quote]I thought we were up to build 3?

[QUOTE=mognuts;164670]I thought we were up to build 3?[/QUOTE]32-bit is on build 3, 64-bit is on build 2.

[QUOTE=James Heinrich;164725]32-bit is on build 3, 64-bit is on build 2.[/QUOTE]No it is build 3 for both 32 and 64 bits : if you look at the properties of the exe included in the compressed file you pointed to, the version is 25.9.3.0. It is also the version reported by Help / About Prime95.

Jacob

No I haven't had time to try any of the Prime95 25.x versions, but I agree that it is a logical next step. Hopefully I will be able to do so during the upcoming weekend. However, since George has mentioned several times that he really hasn't modified the TF code in the Prime95 25.x versions, I remain cautiously optimistic.
I can report another failure (Factor=62019047,65) on a completely different PC using Prime95, version 24.13, (32 bit) and Windows XP SP 2 (32 bit).

[QUOTE=S485122;164735]No it is build 3 for both 32 and 64 bits[/QUOTE]You're right -- apparently I missed the latest x64 version :blush:

This is likely fixed in v25. From the code in commonc.c that generates temporary file names:

[code]/* WARNING: Version 24 had a bug where exponents between 61 million and */
/* 70 million used funky characters in the file name. I've fixed the bug */
/* here, but users will have to rename some intermediate files. */
[/code]

Apparently, the funky characters in the file name are causing a crash (probably from more bad code of mine where errors creating the save file aren't handled properly?).

George is correct about this error being fixed in v25 (verified in 25.9 build 3).
This leads to the following two questions:
1. Can the server be programmed to hand out TF assignments to v24.xx PCs which are [B]outside[/B] of the 61 million to 70 million range? If so, TF assignments greater than 70 million could be efficiently factored to low bit levels by the v24.xx TF PCs that otherwise will stop contributing. I am making this suggestion because I have a few borged 24.xx PCs that are outside of my control, [U]and I am sure that other long time Prime95 contributors are in the same boat.[/U]
2. Are v25 assignments "locked" to the specific PC that registered them with the the V5 server or to the user id associated with that PC? In other words, can I trade v25 assignments among the PCs that I have access to like I can with v24 assignments or do I have to unreserve them? I know how to move the "bad" [U]unreserved[/U] v24 TF assignments from my 24.xx PCs to my v25.9 PCs by using the worktodo.add trick, but I am not sure how to proceed after the TF assignments have been registered with the V5 server.

[QUOTE=RMAC9.5;165432]1. Can the server be programmed to hand out TF assignments to v24.xx PCs which are [B]outside[/B] of the 61 million to 70 million range? [/quote]

I've tried to implement this. It is untested.

[quote]2. Are v25 assignments "locked" to the specific PC that registered them with the the V5 server or to the user id associated with that PC?[/QUOTE]

You can move the assignments to any other PC with the same userid. Be sure to copy that 32 hex digit assignment id too.

George,
I am happy to report that the server is sending out TF assignments greater than 70 million to some of my v24.13 TF PCs. The one 70 million assignment that I checked so far stopped and started normally.

Here are two more v25.9 questions.
1. What does SendAllFactorData=1 in Prime.txt do? All of the PCs that I have direct access to use dial-up modems and will continue to do so for the foreseeable future. If I set it =0, will v25.9 act like v24.13 and send the TF results at the end of the assignment, instead of at the end of each bit level?
2. Does v25.9 put a greater demand on the memory subsystem than v24.13? The reason for this question has to do with my new Phenom II X4 940 PC which happily did DCs at 3.3 GHz using v24.13 but gets "SUM(INPUTS) != SUM(OUTPUTS)" errors for LLs at anything over 3.0 GHz using v25.9.