For fun, have you considered making an mfaktc that could tacke MM128? GPUs should be able to add several more bits to how much trial factoring has been done. You could even start up a mini-project to coordinate the attack.

[QUOTE=firejuggler;251723]4294967295 (yeah, 11 time the billion digit project)[/QUOTE]

Yeah... no. More like 1.3 times (4294967295*log[SUB]10[/SUB]2 = 1,292,913,986 digits)

[QUOTE=firejuggler;251723]a small thing.. i tested mfacktc a bit further and it seems that there is a limit to the exponent you can use. If it is above 4294967295 (yeah, 11 time the billion digit project); it enter the famous loop :
got assignment: exp=4294967295 bit_min=1 bit_max=60
WARNING: exponent is not prime! Ignoring this assignment!

(4294967295=2^32-1)[/QUOTE]
Right, the exponent is limited to 2^32 -1. In the next version ignoring should work. :smile:

[QUOTE=Prime95;251730]For fun, have you considered making an mfaktc that could tacke MM128? GPUs should be able to add several more bits to how much trial factoring has been done. You could even start up a mini-project to coordinate the attack.[/QUOTE]

Not really, would need many changes in the code.

Oliver

It seems that George is showing a lot of interest in GPGPU lately. :smile:

Would it possible to change the interface to present the information on one screen, updating the relevant fields as each class completes, rather than just dumping another line on the screen. Often there are so many lines I can't scroll back far enough to figure out what I'm working on, and it would be nice to have built-in ETA and such. I was thinking something similar to the LAME interface, something like this:[code]mfaktc 0.15 64bits (http://www.mersenneforum.org/showthread.php?t=12827)
CPU features: MMX (ASM used), SSE (ASM used), SSE2
CUDA device info CUDA version info
name GeForce 8800 GT binary compiled for 3.10
compute capability 1.1 CUDA driver version 3.20
max threads/block 512 CUDA runtime version 3.10
num multiprocessors 14 (112 shader cores)
clock rate 1500MHz

Assignment
tf(332222641, 76, 77, ...);
k_min = 113715705073740
k_max = 227431410148576
Using GPU kernel "barrett79_mul32"

Class | Avg. Rate | Avg. Wait | time / estimate / ETA
2900/4620 (62%) | 123.10M/s | 6316us | 1:01:02 / 1:37:14 / 0:36:12[/code]Laid out that way, you'd pretty much just need to overwrite the last line every time, except when a new assignment starts (in which case you need to overwrite the last 7 lines).

Oliver,

Can you write an article for the wiki about mfaktc and pm it to me?
Or post it yourself.

Thanks.

[QUOTE=James Heinrich;252101]Would it possible to change the interface to...
<snip>
[/QUOTE]

Um, don't do that. That makes it impossible to be grep'd or any other scripts. I'm aok, to have an option to create a screen like that. :)

If you want to work out what it's working on, do a "head -1" on the worktodo.txt file.

As for ETA - once you do a couple, you get the hang of how long they take. I created a shell script to time results.txt into a file called results.log. So tail -f on my results.log file I can work out ETAs.

-- Craig

I use on Windows on day Ubuntu on night. Windows's version allow me use my computer while maintaining a 100 - 105 M/s speed (win 32 bit version).
When I use ubuntu (10.10), it seems to hang up often ,but the speed is at 120-125 M/s ( Linux 64 bit version).
Also, it seems that i can't transfert the checkpoint from windows to ubuntu (it seems to ignores it).
So, I would like 2 things :
-A way to tranfers checkpoint
-A way to reduce the priority under linux.

Sorry for my broken english. it's late and I lack cafféine.

[QUOTE=firejuggler;252856]I use on Windows on day Ubuntu on night. Windows's version allow me use my computer while maintaining a 100 - 105 M/s speed (win 32 bit version).
When I use ubuntu (10.10), it seems to hang up often ,but the speed is at 120-125 M/s ( Linux 64 bit version).[/QUOTE]

Seems that you're limited by your CPU on Windows. This is not a Windows issue. The 64bit version is faster than the 32bit version (CPU part).

[QUOTE=firejuggler;252856]Also, it seems that i can't transfert the checkpoint from windows to ubuntu (it seems to ignores it).
So, I would like 2 things :
-A way to tranfers checkpoint
-A way to reduce the priority under linux.

Sorry for my broken english. it's late and I lack cafféine.[/QUOTE]
At the current time there are no priorities in CUDA code possible. You can try to reduce the pressure on your GPU on Linux by setting SievePrimes manually in mfaktc.ini (SievePrimes=N, SievePrimesAdjust=0) a bit higher so that you lower the GPU rate.

Oliver

mfaktc 0.15
 

Hello,

find attached mfaktc 0.15! :smile:
Primary changes:
- one checkpoint file per exponent
- complete rewrite of worktodo parsing, should be more robust now
- added lots of "cudaGetLastError()" in the code, perhaps this gives more information in case of problems

The performance is the same as mfaktc 0.14, the math core is untouched!

I have some Windows binaries on my harddisk (yes, finally I can create them) but I have to verify under what conditions I'm allowed to distribute them. I'm using the "Microsoft Windows SDK for Windows 7 and .NET Framework 3.5 SP1" which includes a 64bit capable compiler for Windows which is supported by CUDA. Any hints on this topic are welcome!

Oliver

Helping PrimeNet in TF
 

I remember I read something about speeding up PrimeNet exponents with low or no P-1 (around M50,000,000) rising their bit levels using TF with CUDA.

Is there a query that can be done to reserve right exponents, maybe taking B1 and B2 fields of P-1 into account? And how many bits may we raise? Is it correct to start from 69 to 71 or 72 bits?

Luigi