I'm trying to get v25.6 to give me P-1 factoring work. I'm using that setting for the "Type of work to get", but the v5 server is giving me Double-check tests instead. This happens when I try to set it on only 1 thread or when I try on "All worker threads".

Any advise would be appreciated.

Thanks.

[QUOTE=monst;139384]I'm trying to get v25.6 to give me P-1 factoring work. I'm using that setting for the "Type of work to get", but the v5 server is giving me Double-check tests instead. This happens when I try to set it on only 1 thread or when I try on "All worker threads".

Any advise would be appreciated.

Thanks.[/QUOTE]

Does prime.log indicate why it is giving you double-check work? My only guess is you've not given prime95 enough memory to use.

At work I have a core2duo with 3 Gb ram. Its windows xp 32bit but should still support up to 2Gb for prime95 to use. I was going to give 2000Mb to prime95 but it turned out 1843Mb was max.

I requested ECM on fermat numbers on one core and trialfactoring to low limits on the other and I got ECM on F26 with B1=50000 and B2=5000000. Every time it tries to start ECM stage2 it says using it will use 1742MB but when the prime95.exe process reaches around those 1740Mb used it quits with the error:

"Instruction at "0x00447a62" refered memory at "0x00000000". Memory could not be written"

[QUOTE=ATH;139588]At work I have a core2duo with 3 Gb ram. Its windows xp 32bit but should still support up to 2Gb for prime95 to use. I was going to give 2000Mb to prime95 but it turned out 1843Mb was max.

I requested ECM on fermat numbers on one core and trialfactoring to low limits on the other and I got ECM on F26 with B1=50000 and B2=5000000. Every time it tries to start ECM stage2 it says using it will use 1742MB but when the prime95.exe process reaches around those 1740Mb used it quits with the error:

"Instruction at "0x00447a62" refered memory at "0x00000000". Memory could not be written"[/QUOTE]

I had similar problems on ECM of F24. Did you try to close and restart Prime95?
I had no problems after a clean restart.

Luigi

[QUOTE=ET_;139590]I had similar problems on ECM of F24. Did you try to close and restart Prime95?
I had no problems after a clean restart.

Luigi[/QUOTE]

Yes, I restarted several times from last save at 98.9% done of stage 1, same thing happened each time. Made a new save at 99.98% of stage1 and restarted from there but still crashed. I successfully ECM on F25 at home with same B1 and B2 using around 1000-1100Mb Ram.

[QUOTE=ATH;139588]I was going to give 2000Mb to prime95 but it turned out 1843Mb was max.[/QUOTE]Sounds vaguely similar to the (unsolved) problems I reported back in [url=http://www.mersenneforum.org/showthread.php?t=9779&page=1#post121605]post #28 of this thread[/url], where trying to allocate more than 1800MB (even though I should theoretically be able on win64 to allocate 3GB+ without problem) would result in Prime95 repeatedly reporting problems and auto-cropping my RAM allocation until it was down to only a few hundred MB. Since I've set it to 1800MB (not 1840MB) it seems to have been behaving itself for the last several month (although I'd give it more RAM to chew on if it could use it without hurting itself).

[quote=James Heinrich;139596]Sounds vaguely similar to the (unsolved) problems I reported back in [URL="http://www.mersenneforum.org/showthread.php?t=9779&page=1#post121605"]post #28 of this thread[/URL], where trying to allocate more than 1800MB (even though I should theoretically be able on win64 to allocate 3GB+ without problem) would result in Prime95 repeatedly reporting problems and auto-cropping my RAM allocation until it was down to only a few hundred MB. Since I've set it to 1800MB (not 1840MB) it seems to have been behaving itself for the last several month (although I'd give it more RAM to chew on if it could use it without hurting itself).[/quote]
I let mprime 25.6 use up to 3480 MiB RAM for P-1 and ECM-F on 64bit Linux with no problems. A Windows specific bug perhaps? Try 640 KB. :wink:

(it was reported, I think)
 

Just so we don't forget to fix this, here's a bug (it was reported before, I think):

1. prerequisites - BIOS and CPU are "smart-multipler" capable. E.g. Gigabyte (put your name here) + Q6600. The idle-to-loaded multiplier range can be 6/9, 6/7, etc. The "smart" logic is usually preset to ON, and some users may not know how to change this and/or may not be in control of BIOS changes.
2. Prime95.exe (e.g. 25.6) starts and immediately measures the [SIZE=2]CpuSpeed[/SIZE] parameter ([U]here[/U]: to fix - it should throw some load, wait a second, two, three? and measure the CpuSpeed), reports it to server, writes it to local.txt
3. Based on the measured CpuSpeed (it could be 6/9th of the real speed or 6/7th, etc, or even 1/26th see below) - the program decides that there's too much work in worktodo.txt and releases some tasks.
4. (possibly a bit later) the program decides to get new assignments. They are not the same, of course.

Result: user annoyed (**), some partial work possibly lost (if they juggled assignments in worktodo.txt). Granted, in most cases the last lines from worktodo.txt worker sections will be released; if they user didn't sort/shuffle them*, they were most probably never touched. Anyway.

[U][I]Variant[/I][/U]: I've seen CpuSpeed to be measured as 100 (!) on an AMD (the true CpuSpeed is 2600). Then the consequences are even more drastic: [B]all[/B] the existing work is immediately released, some trial factoring is assigned instead.

It did happen a couple of times in my practice. I've put a door stop into local.txt - but it's and ugly hack and not for everyone - e.g.
[code][SIZE=2]CpuOverride=1[/SIZE]
[SIZE=2]CpuType=12[/SIZE]
[SIZE=2]CpuSpeed=3204[/SIZE][/code]

--Serge

* I move some jobs lower, if I plan to do their P-1 on another computer; they may have started and done some LL already (because other threads were doing P-1 as well)
** or worse... <joke>some may say that they lost their M45 through that!</joke>

[quote]1. prerequisites - BIOS and CPU are "smart-multipler" capable. E.g. Gigabyte (put your name here) + Q6600. The idle-to-loaded multiplier range can be 6/9, 6/7, etc. The "smart" logic is usually preset to ON, and some users may not know how to change this and/or may not be in control of BIOS changes.[/quote]Not only in BIOS, some OS's have support for controling the frequency scaling through the OS too.

[quote] 2. Prime95.exe (e.g. 25.6) starts and immediately measures the [SIZE=2]CpuSpeed[/SIZE] parameter ([U]here[/U]: to fix - it should throw some load, wait a second, two, three? and measure the CpuSpeed), reports it to server, writes it to local.txt[/quote]I like that idea.

Something else to watch out for is the Linux CPU governor. By default it will ignore idle-priority tasks (like mprime) when selecting the CPU speed.

So you could have the situation that the CPU speed is changing often as other tasks are being used.

But I'm not sure how the current mprime handles this. I have the CPU governor disabled so CPU runs at max freq all the time.

[quote]Result: user annoyed (**), some partial work possibly lost (if they juggled assignments in worktodo.txt). Granted, in most cases the last lines from worktodo.txt worker sections will be released; if they user didn't sort/shuffle them*, they were most probably never touched.[/quote]It looks like mprime doesn't release an exponent if it has some LL work started on it, though.

How come even 64bit version can't use >2gb ram?

I have a Core2Quad with 4Gb ram and Windows XP 64bit and prime95 25.6 64bit. I have like 3.2 Gb ram free so I gave prime95 3000 Mb for a p-1 job, but it couldn't use that much:

Memory allocation error. Day and night memory settings changed to 2400MB and 2400MB.
Memory allocation error. Day and night memory settings changed to 1920MB and 1920MB.
Memory allocation error. Day and night memory settings changed to 1536MB and 1536MB.

[QUOTE=ATH;140444]How come even 64bit version can't use >2gb ram?
...
Memory allocation error. Day and night memory settings changed to 2400MB and 2400MB.
Memory allocation error. Day and night memory settings changed to 1920MB and 1920MB.
Memory allocation error. Day and night memory settings changed to 1536MB and 1536MB.[/QUOTE]I've been plagued with that problem for 6+ months (see posts #189, #257, etc from this thread), waiting (impatiently?) for George to release v25.7 x64-edition that actually works with >1800MB allocated...

Please George, make it work... we don't like wasted resources in our systems :down: