mfaktc: a CUDA program for Mersenne prefactoring

[TheJudger]

Originally Posted by Prime95

I am not a lawyer.

I don't know if you can "undo" GPL on mfaktc and release it as public domain instead.

AFAIK in germany it is not possible to release something as real public domain, there must be a copyright holder...

Duallicense might be an option if needed.

To be honest: those license stuff is not the part which makes fun...

Oliver

[TheJudger]

Originally Posted by Christenson

1) Here is my latest spec. I've decided which new keys will be essential, and which optional.

First (quick) view on your specification, two questions:

Does "UpperLimitIncrease=X" increase the upper limit of the assignment by X or does it increase the upper limit to the "final limit" + X?
E.g. M53.xxx.xxx,55,56 with UpperLimitIncrease=2, will it do M53.xxx.xxx

• from 2^55 to 2^58
• from 2^55 to 2^71 using the list from here: http://mersenne.org/various/math.php

If it is the second case and we don't allow negative values of UpperLimitIncrease than the MinWorkUnitMinutes might not be needed.


Is a MinimumExponent key possible? Using manual form I can try request assignments within a specific range. E.g. currently I'm doing TF on 100M digit numbers.

Oliver

[xilman]

Originally Posted by TheJudger

AFAIK in germany it is not possible to release something as real public domain, there must be a copyright holder...

Duallicense might be an option if needed.

To be honest: those license stuff is not the part which makes fun...

Oliver

Yes and no. As I said, IANAL but the copyright holder, even in Germany, may explicitly state that the intellectual property may be reused in any way whatsoever. For all practical purposes this is the same as releasing it to the public domain.

FWIW, the same is true in the UK, AFAIK.


Paul

[TheJudger]

Hello,

here is mfaktc 0.17!

• Users with "fast CPU + slow GPU" can try to enable AllowSleep in mfaktc.ini. This allows the CPU to sleep instead of running a busyloop if the CPU has to wait for the GPU
• replaced compiletime option "THREADS_PER_GRID_MAX" with the runtime option "GridSize" (mfaktc.ini)
  • maximum "threads per grid" is now 2²⁰ (was 2²¹), default value is the same as before
  • maximum SievePrimes increased from 100.000 to 200.000
  • lower GridSize might increase the usability of your GUI while running mfaktc at the cost of slightly lower performance
• no change in performance compared to 0.16p1

As usual: finish your current assignment and upgrade mfaktc to the new version after that (mfaktc refuses checkpoint files from other versions than itself).

Oliver

P.S. those who had access to the -pre versions: mfaktc 0.17-pre2 is mfaktc 0.17 without changes except the version string.

[TheJudger]

Windows 64bit executable
If it reports a missing cudart64_32_16.dll you can
- install the CUDA toolkit
- put this dll into the mfaktc directory (prefered solution) http://www.mersenneforum.org/showpos...&postcount=632

I tried to include the dll but with standard .zip the filesize is ~285kiB which is above the limit of this forum.

[TheJudger]

Windows 32bit executable
If it reports a missing cudart32_32_16.dll you can
- install the CUDA toolkit
- put this dll into the mfaktc directory (prefered solution) http://www.mersenneforum.org/showpos...&postcount=658

[firejuggler]

32 bit version work flawlessly for now

[Christenson]

Originally Posted by TheJudger

Windows 32bit executable
If it reports a missing cudart32_32_16.dll you can
- install the CUDA toolkit
- put this dll into the mfaktc directory (prefered solution) http://www.mersenneforum.org/showpos...&postcount=658

Hey, time to repost the cudart DLLs!

[firejuggler]

3.20 32 bit version here

[Christenson]

Originally Posted by TheJudger

First (quick) view on your specification, two questions:

Does "UpperLimitIncrease=X" increase the upper limit of the assignment by X or does it increase the upper limit to the "final limit" + X?
E.g. M53.xxx.xxx,55,56 with UpperLimitIncrease=2, will it do M53.xxx.xxx

• from 2^55 to 2^58
• from 2^55 to 2^71 using the list from here: http://mersenne.org/various/math.php

If it is the second case and we don't allow negative values of UpperLimitIncrease than the MinWorkUnitMinutes might not be needed.


Is a MinimumExponent key possible? Using manual form I can try request assignments within a specific range. E.g. currently I'm doing TF on 100M digit numbers.

Oliver

Oliver:
The code does not exist, only the specification. The specification is a "straw man", meaning I am committed to the spirit, but not the letter of it. We can make it do anything we want to program, but I want to keep it simple for the first round. I had no intention of allowing negative increases.

My suggestion here is to encode that list of factoring levels in the code, and key "TakeBeyondRecommendedFactoringLevel=3" for the second purpose, or we can name it something shorter that's not coming to me yet tonight.

A "Minimum Exponent" key is certainly possible, if it's in the server interface somewhere. If it isn't there, I'm sure prime95 will add it for us on request.

I am also not sure I like what I suggested for log file handling. The spirit of it is to both keep a record of what has been turned in and to know what to turn in next. The "results.txt" + "prime.log" solution of prime95 is attractive. Are we going to need to have limits on the file size for the logs, and occasionally chop off the front? Additional keys for this aren't exactly expensive, though I'm not sure how to best tell the filesystem to do this without either having dual log files or doing some copying.

Whatever we do, it has to go into the "readme". I'm not going to remember in six months, and I will curse myself if I forget that step!

[Karl M Johnson]

It works!
GPU usage at 43%.