Brain,

sounds like a good idea, now we need someone with the rights to make a thread sticky and someone you can edit those posts for a longer period. "Normal users" can edit their own posts only within a small time period after the initial post.

example:
What to do with GPUs on primenet?
Nvidia[INDENT]Lucas Lehmer test using CUDALucas[INDENT]URL to CUDALucas
<direct link to latest src>
<direct link to latest binray precompiled for OS1>
<direct link to latest binray precompiled for OS2>
...[/INDENT]Lucas Lehmer test using gpuLucas (if available)
Trial factoring using mfaktc[INDENT]URL to mfaktc thread here in the forum
<direct link to latest src>
<direct link to latest binray precompiled for OS1>
<direct link to latest binray precompiled for OS2>
...[/INDENT][/INDENT]ATI/AMD[INDENT]nothing available, go and write your own code :wink:[/INDENT]
Oliver

Perhaps you should ask whomever the relevent person is for an account on [url=http://www.mersennewiki.org/index.php/Main_Page]Mersenne Wiki[/url] so that you can create and maintain a page there about the current progress.

Also, you should definately PM Xyzzy to ask for a bump in privileges to moderate in this forum.

[QUOTE=ATH;240591]I couldn't compile it with Mingw64, but I succeeded with Visual Studio, and it passes all selftests.

I tried to remove the line:
[CODE]else if(!isprime(exp)) printf("WARNING: exponent is not prime! Ignoring this assignment!\n");[/CODE]but it crashes when I try to factor a composite exponent, specifically M(p[sup]2[/sup]).[/QUOTE]

Same issue here... division by zero exception. Not sure if it a bug in the program or the fact I hacked the makefile to complie for win32 - the makefile for windows seems to be for x64. In any case the self tests pass and factorization on runs with no exceptions for prime p.

As a new mfaktc user, a couple comments (i7-920, 8800GT, Win7x64):
* I'd much prefer the output lines to say "2345/4620" instead of "class 2345" to give some idea of progress (it's far from obvious how many classes are to be processed)
* worktodo validation should be a little more robust -- a Factor line without an assignment hash crashes mfaktc rather than providing a warning about an invalid assignment
* is it possible that the savefile could hold progress for more than one test, for example if I wanted to suspend work on the current assignment and work on something else. Of course it can be worked around manually (multiple instances in different directories, backup copies, etc) but the contents of mfaktc.ckp seem very well suited to holding more than one test

[QUOTE=James Heinrich;244941]
* I'd much prefer the output lines to say "2345/4620" instead of "class 2345" to give some idea of progress (it's far from obvious how many classes are to be processed)[/QUOTE]

OK, with removing the word "class" it should fit to the 80 culumn limit.

[QUOTE=James Heinrich;244941]
* worktodo validation should be a little more robust -- a Factor line without an assignment hash crashes mfaktc rather than providing a warning about an invalid assignment[/QUOTE]

I'll look at it, did you write "Factor=,<exp>,<bit_min>,<bit_max>" or "Factor=<exp>,<bit_min>,<bit_max>" into the worktodo?

[QUOTE=James Heinrich;244941]
* is it possible that the savefile could hold progress for more than one test, for example if I wanted to suspend work on the current assignment and work on something else. Of course it can be worked around manually (multiple instances in different directories, backup copies, etc) but the contents of mfaktc.ckp seem very well suited to holding more than one test[/QUOTE]

Is this really a useful feature?

Oliver

I just had "Factor=<exp>,<bit_min>,<bit_max>" in worktodo. Maybe accepting any of the 3 variants would be fine (/^Factor=([0-9A-Z]{32})?,?([0-9]{7,9}),([0-9]{2}),([0-9]{2})$/) but the important thing is to ONLY accept the line if it matches all components (exp, bit_min, bit_max all within expected ranges) and not attempt to blindly accept any line beginning with "Factor=".

I don't know how useful multi-job savefiles are overall, but I know just playing around I lost 8 hours of overnight work because I started a different (much smaller) assignment briefly, not realizing it would clobber all the work I did on the previous one. I figured it would be a relatively simple extension to loop through multiple lines in the savefile rather than just taking the first/only one.

[QUOTE=James Heinrich;244972]I don't know how useful multi-job savefiles are overall, but I know just playing around I lost 8 hours of overnight work because I started a different (much smaller) assignment briefly, not realizing it would clobber all the work I did on the previous one. I figured it would be a relatively simple extension to loop through multiple lines in the savefile rather than just taking the first/only one.[/QUOTE]

Or... just do what Prime95/mprime does, and have a separate save file for each exponent which has been worked on but not yet completed.

Not complicated to implement if the save file's file-name is derived from the exponent....

[QUOTE=chalsall;244977]Or... just do what Prime95/mprime does, and have a separate save file for each exponent which has been worked on but not yet completed.

Not complicated to implement if the save file's file-name is derived from the exponent....[/QUOTE]

added to .plan for mfaktc 0.15 aswell as the problems you've noticed on worktodo parsing.

Oliver

[QUOTE=James Heinrich;244941]
* I'd much prefer the output lines to say "2345/4620" instead of "class 2345" to give some idea of progress (it's far from obvious how many classes are to be processed)[/QUOTE]

I had the exact same thought yesterday:

[CODE]--- mfaktc-0.13p1/src/tf_common.cu 2010-12-05 17:29:46.000000000 -0500
+++ ../mfaktc-0.13p1/src/tf_common.cu 2011-01-06 21:07:11.964571175 -0500
@@ -290,7 +290,8 @@

if(mystuff->mode != MODE_SELFTEST_SHORT)
{
- printf("class %4" PRIu64 ": ", k_min%NUM_CLASSES);
+ printf("class %4" PRIu64 "/%4d (%.2f%%): ", k_min%NUM_CLASSES, NUM_CLASSES,
+ ((double)(k_min%NUM_CLASSES)) * 100 / NUM_CLASSES);
if(((unsigned long long int)mystuff->threads_per_grid * (unsigned long long int)count) < 1000000000ULL)
printf("%.2fM", (double)mystuff->threads_per_grid * (double)count / 1000000.0f);
else
[/CODE]Outputs as:
[QUOTE]class 349/ 420 (83.10%): 191.89M candidates in 3618ms (53.04M/sec) (avg. wait: 279usec)[/QUOTE]

(on a side note, I found it to run a little faster without MORE_CLASSES, on my gts250. Is this right?)

[QUOTE=James Heinrich;244972]
I don't know how useful multi-job savefiles are overall, but I know just playing around I lost 8 hours of overnight work .....[/QUOTE]

Out of curiosity what work took 8hrs?

Longest job I've seen is under 90mins on a GTX460. For me save files are almost irrelevant.

As a work around you could rename mfaktc.ckp if you change work.

-- Craig

I don't remember exactly what the job was, but I believe it was something like 2^76-2^77 on something in the M333M range. My 8800 isn't as fast as your 460 ;)