mfaktc: a CUDA program for Mersenne prefactoring

[TheJudger]

Hi Jason,

Originally Posted by jasonp

Maybe you could include the same kernel compiled two different ways, then choose which one to use based on the number of GPU registers as reported by the Nvidia driver.

I would say that this is not needed. The 71bit kernel uses 16 registers with and without the limit to 16 registers.
The 75bit kernel uses 17/16 (limit off / on) and the 95bit kernel uses 18/16 registers. 17 or 18 is basically the same since registers are allocated in groups of 4.
I made some tests on my GTX 275 (compute capability 1.3, 16384 register per multiprocessor) and didn't notice any difference in performance. With and without limit to 16 registers it runs a the same speed.
On my 8400GS (compute capability 1.1, 8192 register per multiprocessor) it runs ~1% faster with the limit to 16 registers! It think this is related to the fact that the occupancy is higher.
16 registers * 256 threads per block = 4096 registers per block ==> two blocks can run at the same time one the same multiprocessor!
(With 192 threads per block I could use 20 registers per block an run 2 blocks at the same time...)

[TheJudger]

Originally Posted by TheJudger

(With 192 threads per block I could use 20 registers per block an run 2 blocks at the same time...)

less than 0.1% faster than 256 threads per block and limit to 16 registers on GS8400. Definitely not worth the extra work/code.

[TheJudger]

Hi David,

Originally Posted by henryzz

Currently at OBD all the available assignments are taking numbers on from 75 bits or more. Based on testing upto 70 bits 75-76 will take me ~8.4 hours. I can't often guarantee that my pc will be running that long at once but I would like to help out a bit. Is there any chance of making partial bit levels available or having some sort of saving feature.

mfaktc 0.08 has resume capability. Release is planned for the next few days.

Oliver

[henryzz]

Originally Posted by TheJudger

Hi David,



mfaktc 0.08 has resume capability. Release is planned for the next few days.

Oliver

Brilliant news.
I wait expectantly.

[Karl M Johnson]

How do I force Prime95 to bench a exponent with fixed bounds like bit_min and bit_max similar to mfaktc ?
How should worktodo.txt file look like inside?

[ET_]

Originally Posted by Karl M Johnson

How do I force Prime95 to bench a exponent with fixed bounds like bit_min and bit_max similar to mfaktc ?
How should worktodo.txt file look like inside?

worktodo.txt:
Factor=bla,exponent,bitmin,bitmax

Luigi

[Karl M Johnson]

Prime95 doesnt like that bla. I assume feeding a random hash of required length will calm it. Here's an example from PrimeNET : hash,49653607,69,0. Now, why is bitmin 69 and bitmax 0 ? Is bitmax = 0 = infinity ?

[Mini-Geek]

Originally Posted by Karl M Johnson

Prime95 doesnt like that bla. I assume feeding a random hash of required length will calm it.

If you don't have an assignment key (that's what's supposed to go in place of the "bla"), leave it blank with no leading comma, (Factor=exponent,bitmin,bitmax) or put "N/A" (Factor=N/A,exponent,bitmin,bitmax).

Originally Posted by Karl M Johnson

Here's an example from PrimeNET : hash,49653607,69,0. Now, why is bitmin 69 and bitmax 0 ? Is bitmax = 0 = infinity ?

Is that example line from an LL test or DC and not a TF assignment? For Test= and DoubleCheck=, the last part there (the ",0") isn't bitmax, it's has_been_pminus1ed (1 if the number has had a P-1 run, 0 if it hasn't).

[chalsall]

Originally Posted by Karl M Johnson

Prime95 doesnt like that bla. I assume feeding a random hash of required length will calm it. Here's an example from PrimeNET : hash,49653607,69,0. Now, why is bitmin 69 and bitmax 0 ? Is bitmax = 0 = infinity ?

Personally, I use "DEADBEEFDEADBEEFDEADBEEFDEADBEEF" for "bla". It's legitimate hexadecimal.

[TheJudger]

Hi,

find attached mfaktc 0.08.

Highlights:
- 2 new GPU kernels for factors up to 2^75 and 2^95 (above 2^90 isn't tested very well )
- resume capability

For details take a look at Changelog.txt and README.txt.

Thank you Luigi (ET_) and Kevin (kjaget) for testing and comments!

Oliver

[TheJudger]

Hi Karl,

Originally Posted by Karl M Johnson

Prime95 doesnt like that bla. I assume feeding a random hash of required length will calm it. Here's an example from PrimeNET : hash,49653607,69,0. Now, why is bitmin 69 and bitmax 0 ? Is bitmax = 0 = infinity ?

I think you have to override the factor defaults. Prime95 automatically sets the upper limit unless you override it. I would take a look into undoc.txt and search for "factor override" in the forum/web.

Oliver