How to run MFAKTC in Linux?

[chris2be8]

Originally Posted by Runtime Error

Yes, I mentioned doing that in my original post on this thread.

To solve my issue, it seems like following the steps here would work, but I do not have super-user privileges. https://unix.stackexchange.com/quest...-in-centos-6-5

It says that you can install 2.14 alongside 2.12 easily, then use it to compile projects with the following steps:

mkdir ~/glibc_install; cd ~/glibc_install 

wget http://ftp.gnu.org/gnu/glibc/glibc-2.14.tar.gz

tar zxvf glibc-2.14.tar.gz

cd glibc-2.14

mkdir build

cd build

../configure --prefix=/opt/glibc-2.14

make -j4

sudo make install

export LD_LIBRARY_PATH="/opt/glibc-2.14/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}"

Is there a way that I can somehow get around this? Thanks in advance.

You should be able to run everything before sudo make install under your account. Then export LD_LIBRARY_PATH="/home/chris/glibc_install/glibc-2.14;$LD_LIBRARY_PATH" replacing /home/chris with the full path to your home dir should work. You may need to tweak the path to be the dir where libc.so.6 is.

Or need to change ../configure --prefix=/opt/glibc-2.14 to ../configure --prefix=/path/to/current/dir

Chris

[paulunderwood]

I downloaded Centos 6, installed it virtually, updated it and am now downloading the CUDA toolkit. Expect a binary executable soon.

[paulunderwood]

Hello from virtual Centos 6!

Please try the attached programs for glibc-2.12. If they fail to run, please let us know.

[Runtime Error]

Thank you so much, I will try these as soon as I can. I appreciate the help!

[Runtime Error]

Thank you so much for taking the time to compile these, and thanks for helping me learn how to do this kind of stuff! I appreciate your efforts.

Unfortunately the new error is, I think, because the latest installed CUDA module on this system is older, and this file you generously shared was built on 10.2. Here is the error:

CUDA version info
binary compiled for CUDA 10.20
CUDA runtime version 4414.65
CUDA driver version 10.10
ERROR: current CUDA driver version is lower than the CUDA toolkit version used during compile!
Please update your graphics driver.

And here are the listed CUDA modules

$ module avail
cuda/cuda50/blas/5.0.35
cuda/cuda50/fft/5.0.35
cuda/cuda50/profiler/5.0.35
cuda/cuda50/tdk/5.319.43
cuda/cuda50/toolkit/5.0.35
cuda/cuda55/toolkit/5.5
cuda/cuda65/blas/6.5
cuda/cuda65/fft/6.5
cuda/cuda65/toolkit/6.5
cuda/cuda70/openmpi/gcc/1.10.2
cuda/cuda70/openmpi/openacc/1.10.2
cuda/cuda70/toolkit/7.0
cuda/cuda75/mvapich2/gcc/2.2b
cuda/cuda75/mvapich2/openacc/2.2b
cuda/cuda75/openmpi/gcc/1.10.2
cuda/cuda75/openmpi/openacc/1.10.2
cuda/cuda75/toolkit/7.5
cuda/cuda80/toolkit/8.0

Also, I tried Chris's suggestion above and it gives the following:

CUDA version info
binary compiled for CUDA 6.50
CUDA runtime version 0.1
CUDA driver version 10.10
ERROR: CUDA runtime version must match the CUDA toolkit version used during compile!

My suspicion is that I am not loading something correctly. If so, what might I try?. Any advice is welcome. Thank you again in advance!

[paulunderwood]

Tha attached ought to work for you. Let us know if they work or not

[Runtime Error]

Thank you for the new programs. I think it is very close to working. Definitely my fault for not providing information sooner. It seems that the most recent CUDA support files installed here are "cuda/10.0" when I run "$ module spider cuda" as shown below. If that is correct, I'm confused because I thought the error was because this system has driver version 10.10, also shown below. However, attempting to load "cuda/10.1" gives a module unknown error. Also available is "cuda-supplement/10.1" which requires "pgi/19" but loading these does not alter the below error message.

Thank you for your patience!

CUDA version info
binary compiled for CUDA 10.10
CUDA runtime version 4414.65
CUDA driver version 10.10
ERROR: CUDA runtime version must match the CUDA toolkit version used during compile!

Which is different from last time: "ERROR: current CUDA driver version is lower than the CUDA toolkit version used during compile! Please update your graphics driver."

$ module spider cuda
----------------------------------------
cuda:
----------------------------------------
Description:
The CUDA Support Files
Versions:
cuda/7.5
cuda/8.0
cuda/9.0
cuda/9.1
cuda/9.2
cuda/10.0
Other possible modules matches:
cuda-supplement pycuda scikit-cuda

$module list
Currently Loaded Modules:
1) slurm/18.08 2) xalt/2.7 3) binutils/2.32 4) gcc/6 5) cuda/10.0

$ module load cuda/10.1
Lmod has detected the following error: The following module(s) are unknown: "cuda/10.1"

Follow up question: What is the difference between the three files (3.0, 3.5, 5.0) that you posted? All three give the same errors currently. Thanks.

[paulunderwood]

The 3 files reflect the architecture: https://en.wikipedia.org/wiki/CUDA#GPUs_supported

I am downloading cuda toolkit 10.0 now. Let's hope that works

This is all pretty new to me. Thanks for your informative outputs.

Perhaps somebody here can explain the difference between modules, runtime, drivers and toolkit versions.

[VBCurtis]

This exchange is useful to me, too!
I got stuck in CUDA-version-hell at some point with msieve-GPU, getting errors like these when I accidentally allowed ubuntu to update one of the packages. If y'all find success, I may just have a go trying to get all my CUDA version numbers to match in the hopes of getting msieve or GPU-ECM working again!
Thanks for your efforts, Paul. :)

[paulunderwood]

These may be worth a try, but I am not hopeful.

[paulunderwood]

I am doing a fresh install of Centos 6 virtually with Qemu. I will install Cuda 10.1 on a copy of it. Then compile mfaktc.

I think the problem has been that there are still bits of the 10.2 driver hanging about, if not stubs. Once we have something that is working I will remove all previous compilations from this thread.