How to run MFAKTC in Linux?

[Rodrigo]

Hi Oliver,

MFAKTC 0.21 failed with the following output:

  binary compiled for CUDA  6.50
  CUDA runtime version      -23603.-4
  CUDA driver version       7.50

Hmm, a negative decimal fraction of a decimal number...

And here is the output of the two commands that you requested:

For ldd ./mfaktc.exe --

linux-vdso.so.1 =>  (0x00007ffe0e3e8000)
libcudart.so.6.5 => not found
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f3748e13000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f3748b03000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f37488ed000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f3748523000)
/lib64/ld-linux-x86-64.so.2 (0x00007f3749147000)

For LD_LIBRARY_PATH="./lib:${LD_LIBRARY_PATH}" ldd ./mfaktc.exe --

linux-vdso.so.1 =>  (0x00007ffcb3bb7000)
libcudart.so.6.5 => ./lib/libcudart.so.6.5 (0x00007f62c271f000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f62c23ed000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f62c20dd000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f62c1ec7000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f62c1afd000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f62c18f8000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f62c16da000)
librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f62c14d2000)
/lib64/ld-linux-x86-64.so.2 (0x00007f62c2971000

Hope this will help.

Rodrigo

[Rodrigo]

Doctor, what's the diagnosis?

Rodrigo

[Mark Rose]

What's the output of `lsmod | grep nv` ?

Also try running mfaktc.exe as root at least once after every boot. I have to do that.

[TheJudger]

Rodrigo: as suggested: try to run as root if possible.

Rodrigo/Mark: did you try to run e.g. 'nvidia-smi' as root. Check /dev/nvidia* before and after. Might need to configure udev or whatever to create device nodes and/or set access rights.

Oliver

[Mark Rose]

I do have to create the device nodes if the nvidia module is not loaded at boot. I have one machine where I use the Intel graphics so the desktop stays snappy. I have a script that makes them.

On the other machines I believe it's some kind of permission issue. Running any cuda app as root fixes it. I haven tracked down the exact change being made.

[Rodrigo]

Originally Posted by Mark Rose

What's the output of `lsmod | grep nv` ?

Also try running mfaktc.exe as root at least once after every boot. I have to do that.

Here's the result from the command you requested:

nvidia               8646656  71 
drm                   344064  3 nvidia

But we have great news from running MFAKTC.EXE as root: the self-test [-st] is running as I type this!

Thank you very much!



Rodrigo

UPDATE: The self-test passed, but now I can't get MFAKTC.EXE to run again and I can't remember how I did it the first time. Rebooting didn't help.

[Rodrigo]

UPDATE 2: I figured out what I was doing wrong. The self-test ran fine again, and now MFAKTC.EXE is running an actual TF assignment.

Very cool. This is a major step for me toward making the switch over to full-time Linux and away from Microsoft's Windows 10 snooping and forced bundled updates.

Rodrigo

[Runtime Error]

I'm resurrecting this old thread because it helped me overcome the same issue the original poster was having, and I'd like to follow up with a question of my own. (Warning: noob here)

Currently, I'm seeing the following error when I attempt to run mfaktc on a GPU on a Linux cluster:

/myfolderpath/mfaktc.exe: /lib64/libc.so.6: version `GLIBC_2.14' not found (required by /myfolderpath/mfaktc.exe)

Stack Exchange says this means the program was compiled against glibc version 2.14, and it requires that version to run, but the system I'm using has an older version installed. So need to either recompile the program against the version of glibc that's on my system, or install a newer version of glibc.

However, I don't know how to do either of these. Is it possible that I only need to find "libc.so.6" somewhere and then add a library path to it? Or can I somehow recompile mfaktc using my system's version of glibc?

I'm calling mfaktc from a bash file, if that information helps. And thank you very much in advance for any help!!!

[paulunderwood]

Originally Posted by Runtime Error

I'm resurrecting this old thread because it helped me overcome the same issue the original poster was having, and I'd like to follow up with a question of my own. (Warning: noob here)

Currently, I'm seeing the following error when I attempt to run mfaktc on a GPU on a Linux cluster:



Stack Exchange says this means the program was compiled against glibc version 2.14, and it requires that version to run, but the system I'm using has an older version installed. So need to either recompile the program against the version of glibc that's on my system, or install a newer version of glibc.

However, I don't know how to do either of these. Is it possible that I only need to find "libc.so.6" somewhere and then add a library path to it? Or can I somehow recompile mfaktc using my system's version of glibc?

I'm calling mfaktc from a bash file, if that information helps. And thank you very much in advance for any help!!!

Please post the output form the command uname -a. If is Ubuntu/Debian follow these two posts:

https://www.mersenneforum.org/showpo...postcount=3233

https://www.mersenneforum.org/showpo...postcount=3237

Note I meant "synaptic" rather than "aptitude".

[Runtime Error]

Thanks for the quick reply.

Originally Posted by paulunderwood

Please post the output form the command uname -a.

Linux theclustername 2.6.32-754.23.1.el6.x86_64 #1 SMP Tue Sep 17 09:46:55 EDT 2019 x86_64 x86_64 x86_64 GNU/Linux

[paulunderwood]

Originally Posted by Runtime Error

Thanks for the quick reply.



Linux theclustername 2.6.32-754.23.1.el6.x86_64 #1 SMP Tue Sep 17 09:46:55 EDT 2019 x86_64 x86_64 x86_64 GNU/Linux

This looks like Centos to me. You can follow the instructions in the links I just gave replacing the "apt" stuff with the appropriate "yum" commands.