I have no idea of how to get this working

[bluecat0531]

So ive tried to follow the available guides and tried searching through various forum posts for hours but to no avail. I have a 5950x that im happily running prime95 on but i want to set up my 3090 to do trial-factoring. I believe i have setup MISFIT/GPU72 to get and turn in my work but i cant get either mfaktc or mfaktx controller to get going. Pls help.

[Uncwilly]

Welcome to the forum.
Here is a pointer to a reference guide. https://www.mersenneforum.org/showthread.php?p=488518
Someone with more experience can help answer questions that you have.

[kriesel]

Originally Posted by bluecat0531

I believe i have setup MISFIT/GPU72 to get and turn in my work but i cant get either mfaktc or mfaktx controller to get going. Pls help.

Log everything. Read the logs for clues. Divide and conquer; first get mfaktc working alone, without Misfit/gpu72. If you still can't figure it out, show us some logs. As things stand now, we don't know what issues you're having, because you've provided almost zero information. (Did mfaktc run at all? Did it pass the selftest? Is there anything in the worktodo file? Does the worktodo file even exist? Where did you install it? What error messages or console output do you get when you try to run mfaktc? Misfit? Etc, etc, etc.)

[bluecat0531]

Originally Posted by kriesel

Log everything. Read the logs for clues. Divide and conquer; first get mfaktc working alone, without Misfit/gpu72. If you still can't figure it out, show us some logs. As things stand now, we don't know what issues you're having, because you've provided almost zero information. (Did mfaktc run at all? Did it pass the selftest? Is there anything in the worktodo file? Does the worktodo file even exist? Where did you install it? What error messages or console output do you get when you try to run mfaktc? Misfit? Etc, etc, etc.)

So i cant figure out how to run mfaktc at all. Ive tried using a cmdline but i dont know any commands at all and trying a batch file was no help either, my only other option was just double clicking the .exe file and all that happens is a cmdline appears for one second then disappears.

[chris2be8]

What operating system are you using?

Where have you installed mfaktc?

Open a terminal window, cd into wherever you have installed mfaktc and run if from there. (Under Linux or other UNIX variant type ./mfaktc and press Enter.) Then post *all* the output you get from it here.

[Uncwilly]

Do you have a worktodo.txt established? And IIRC there is a need for a config file to be established before it will stay running. That part was confusing to me too. (I tried running it on a machine with an integrated GPU. But it used shared memory and lowered my desired Prime95 performance.)

[kriesel]

Until the OP posts some useful detail of what he's tried and what it produced, makes some effort to provide data about what's occurring, I suggest we all stop guessing in the dark and trying to help. Our effort should not exceed his.

[Uncwilly]

My answer was commensurate with the OP's most recent post.

[kriesel]

Originally Posted by Uncwilly

the OP's most recent post.

I don't have an issue with your posts or Chris's. Look again at post 4. He quoted my entire post, that included several questions, answered none, and provided little information or specifics. We don't know if he has installed mfaktc in Program Files or system32 or somewhere sensible, has a worktodo file, has proper permissions on the install folder and mfaktc files, has a suitable CUDA computing compatible driver installed, what batch file he has attempted, what cmd line he has attempted, etc etc etc. The details which we would need to see to provide effective help are all being kept secret. I think we may have another "help me in spite of my refusing to provide any useful diagnostic data or answer any questions" troll.

[Viliam Furik]

Originally Posted by bluecat0531

So i cant figure out how to run mfaktc at all. Ive tried using a cmdline but i dont know any commands at all and trying a batch file was no help either, my only other option was just double clicking the .exe file and all that happens is a cmdline appears for one second then disappears.

You need to make a text file called worktodo.txt, where you will have the work for the mfaktc program. It is in the following format:

Factor=[exponent],[bitlevel-start],[bitlevel-end]

Example:
Factor=500000071,82,83

This example will search through all potential factors between 2^82 and 2^83. You can run it as a test, and it should take about 11 hours - in case you finish it, it will output a line to a text file "results.txt", which you should beforehand create as an empty file in case it doesn't do it automatically. There are of course other exponents to do TF (trial factoring) on, and those needed the most (above 120M exponents) take at most 10 minutes with your GPU, as they currently need TF from 2^74 to 2^75.

This should be all to make it work as to having the worktodo file.

It's however almost certain you don't have the necessary version of mfaktc, as I too didn't find a version supporting 3090 (I have 3080), but I didn't ask about it yet. You will also need a cudart64_(version number).dll - the number is most probably 104 or 105, as it mentions it needs mfaktc compiled for CUDA 10.4 or 10.5. You might also need to install the CUDA files necessary for running programs on GPU.


That's about it. If you have any questions, do ask.

[kriesel]

Originally Posted by Viliam Furik

You need to make a text file called worktodo.txt, where you will haveinsert the work for the mfaktc program. It is in the following format:

Factor=[exponent],[bitlevel-start],[bitlevel-end]

Example:
Factor=500000071,82,83

This example will search through all potential factors between 2^82 and 2^83. You can run it as a test, and it should take about 11 hours - in case you finish it, it will output a line to a text file "results.txt", which you should beforehand create as an empty file in case it doesn't do it automatically. There are of course other exponents to do TF (trial factoring) on, and those needed the most (above 120M exponents) take at most 10 minutes with your GPU, as they currently need TF from 2^74 to 2^75.
...

It's however almost certain you don't have the necessary version of mfaktc, as I too didn't find a version supporting 3090 (I have 3080), but I didn't ask about it yet. You will also need a cudart64_(version number).dll - the number is most probably 104 or 105, as it mentions it needs mfaktc compiled for CUDA 10.4 or 10.5. You might also need to install the CUDA files necessary for running programs on GPU.

In all my years of running numerous CUDA versions of mfaktc on multiple GPUs, I don't recall ever needing to create results.txt. Checking in the source code shows it uses an append option call of fopen to write results lines.
https://www.cplusplus.com/reference/cstdio/fopen/ states the file will be created if it does not exist. In module output.c:

void print_result_line(mystuff_t *mystuff, int factorsfound)
/* printf the final result line (STDOUT and resultfile) */
{
  char UID[110]; /* 50 (V5UserID) + 50 (ComputerID) + 8 + spare */
  char string[200];
  
  FILE *resultfile=NULL;
   
  
  if(mystuff->V5UserID[0] && mystuff->ComputerID[0])
    sprintf(UID, "UID: %s/%s, ", mystuff->V5UserID, mystuff->ComputerID);
  else
    UID[0]=0;
    
  if(mystuff->mode == MODE_NORMAL)
  {
    resultfile = fopen(mystuff->resultfile, "a");
    if(mystuff->print_timestamp == 1)print_timestamp(resultfile);
  }
  if(factorsfound)
  {
#ifndef MORE_CLASSES
    if((mystuff->mode == MODE_NORMAL) && (mystuff->stats.class_counter < 96))
#else
    if((mystuff->mode == MODE_NORMAL) && (mystuff->stats.class_counter < 960))
#endif
    {
      sprintf(string, "found %d factor%s for %s%u from 2^%2d to 2^%2d (partially tested) [mfaktc %s %s]", factorsfound, (factorsfound > 1) ? "s" : "", NAME_NUMBERS, mystuff->exponent, mystuff->bit_min, mystuff->bit_max_stage, MFAKTC_VERSION, mystuff->stats.kernelname);
    }
    else
    {
      sprintf(string, "found %d factor%s for %s%u from 2^%2d to 2^%2d [mfaktc %s %s]", factorsfound, (factorsfound > 1) ? "s" : "", NAME_NUMBERS, mystuff->exponent, mystuff->bit_min, mystuff->bit_max_stage, MFAKTC_VERSION, mystuff->stats.kernelname);
    }
  }
  else
  {
    sprintf(string, "no factor for %s%u from 2^%d to 2^%d [mfaktc %s %s]", NAME_NUMBERS, mystuff->exponent, mystuff->bit_min, mystuff->bit_max_stage, MFAKTC_VERSION, mystuff->stats.kernelname);
  }

  if(mystuff->mode != MODE_SELFTEST_SHORT)
  {
    printf("%s\n", string);
  }
  if(mystuff->mode == MODE_NORMAL)
  {
    fprintf(resultfile, "%s%s\n", UID, string);
    fclose(resultfile);
  }
}

RTX30xx are compute capability 8.6. That will require a compile with & dlls for CUDA 11.x. https://en.wikipedia.org/wiki/CUDA Download mirror at https://download.mersenne.ca/mfaktc/mfaktc-0.21 Go for the 2047M GpuSieveSize capable version to get full performance from these fast GPUs.
CUDA 10.x is required for RTX20xx & GTX16xx.
MacOS support has been dropped after CUDA 10.2.