Installation and running of mkfactc

[chalsall]

Originally Posted by daxmick

Do you have source code? The link http://www.mersenneforum.org/mfaktc/ doesn't provide source (and it's the only link I know of for getting mfaktc code).

We cross-posted. The source code is in http://www.mersenneforum.org/mfaktc/...tc-0.21.tar.gz

[daxmick]

Originally Posted by chalsall

We cross-posted. The source code is in http://www.mersenneforum.org/mfaktc/...tc-0.21.tar.gz

Thank you! It would be nice if this URL was in the Sticky post for mfaktc, FYI.

[TheJudger]

For the precompiled binaries there is no need to have CUDA toolkit installed. All you need is a sufficient driver version. Is the CUDA runtime library in the lib/ subdirectory?
What GPU do you have? That binary won't work with Pascal generation carfs (e.g. Geforce 10x0) because of the old CUDA version.

If you have CUDA toolkit already installed: it is not a big deal to compile the sources.
Just edit the Makefile:

• remove/comment the two lines with "--generate-code arch=compute_11,code=sm_11" and "--generate-code arch=compute_20,code=sm_20"
• add a line with "--generate-code arch=compute_60,code=sm_60" for Pascal generation cards

And run

make

Oliver

[daxmick]

Originally Posted by chalsall

We cross-posted. The source code is in http://www.mersenneforum.org/mfaktc/...tc-0.21.tar.gz

I downloaded, extracted the source. I tried to try to "make" but got errors.

daxm@ultraman:/usr/src/mfaktc-master/src$ make
gcc -Wall -Wextra -O2 -I/usr/local/cuda/include/ -malign-double -funroll-all-loops -c sieve.c -o sieve.o
sieve.c: In function ‘sieve_init_class’:
sieve.c:202:23: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
     if((prime_base[j] != exp) && \
                       ^~
sieve.c:248:33: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
     if(k < 0 || k >= p || check != (p-1))
                                 ^~
sieve.c: In function ‘sieve_candidates’:
sieve.c:400:11: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
     for(;i<(SIEVE_SIZE&0xFFFFFFE0) && k<(ktab_size-33);i+=32) // thirty-three!!!
           ^
gcc -Wall -Wextra -O2 -I/usr/local/cuda/include/ -malign-double -c timer.c -o timer.o
gcc -Wall -Wextra -O2 -I/usr/local/cuda/include/ -malign-double -c parse.c -o parse.o
gcc -Wall -Wextra -O2 -I/usr/local/cuda/include/ -malign-double -c read_config.c -o read_config.o
read_config.c:22:10: fatal error: cuda_runtime.h: No such file or directory
 #include <cuda_runtime.h>
          ^~~~~~~~~~~~~~~~
compilation terminated.
Makefile:58: recipe for target 'read_config.o' failed
make: *** [read_config.o] Error 1

I did a search of my filesystem for cuda_runtime.h and found nothing.
I'm guessing that mfaktc needs updated to support CUDA 9?

[daxmick]

Originally Posted by TheJudger

For the precompiled binaries there is no need to have CUDA toolkit installed. All you need is a sufficient driver version. Is the CUDA runtime library in the lib/ subdirectory?
What GPU do you have? That binary won't work with Pascal generation carfs (e.g. Geforce 10x0) because of the old CUDA version.

If you have CUDA toolkit already installed: it is not a big deal to compile the sources.
Just edit the Makefile:

• remove/comment the two lines with "--generate-code arch=compute_11,code=sm_11" and "--generate-code arch=compute_20,code=sm_20"
• add a line with "--generate-code arch=compute_60,code=sm_60" for Pascal generation cards

And run

make

Oliver

I cross-posted. Let me try to answer this now. :-)

I have a GeForce GTX950 (?) I think. Maybe it is a 850. I'll have to double check.

[chalsall]

Originally Posted by daxmick

I downloaded, extracted the source. I tried to try to "make" but got errors. I did a search of my filesystem for cuda_runtime.h and found nothing.

Thank you for your perseverance. In all honesty, I'm at a bit of a loss. This should be easier.

I ran this query on one of my machines, and got this result:

[chalsall@samba2 mfaktc]$ rpm -qf /usr/local/cuda-7.5/targets/x86_64-linux/include/cuda_runtime.h
cuda-cudart-dev-7-5-7.5-18.x86_64

Does that help at all?

[daxmick]

Originally Posted by chalsall

Thank you for your perseverance. In all honesty, I'm at a bit of a loss. This should be easier.

It should be I thanking you all for helping me! I REALLY want to get this to work and I'm sorry that I'm being such a problem. Though I work in Linux every day I don't build from scratch that often so my "skillz" are a bit rusty.

I'm going to go back to basics here and try to re-install the CUDA SDK and drivers and then try again.

All told I *think* I'm getting closer! :-)

[daxmick]

Originally Posted by daxmick

I cross-posted. Let me try to answer this now. :-)

I have a GeForce GTX950 (?) I think. Maybe it is a 850. I'll have to double check.

I have a GeForce GTX 950:

daxm@ultraman:~/Downloads$ lspci | grep -i nvidia
01:00.0 VGA compatible controller: NVIDIA Corporation GM206 [GeForce GTX 950] (rev a1)
01:00.1 Audio device: NVIDIA Corporation Device 0fba (rev a1)

[daxmick]

Woot! Success! The "-st" option is now running TF!

What did I do?
1. I followed the instructions to install CUDA 9.1: https://developer.nvidia.com/cuda-do..._type=deblocal
2. I rebooted.

Honestly I think it was #2 that did the trick. Prior to rebooting I issued the command 'nvidia-smi' and got an error. After reboot that command started working again.

Time to go get some Assignments! I'm so excited!

[daxmick]

Hi HO. hi ho. Hi Ho, Hi Ho; its off to work I go...

[daxmick]

Am I getting too excited over this? I've never done anything "GPU computing" related before.

A HUGE thank you to all of you who helped me work through my woes.