problem installing CUDA on Debian

ixfd64 · 2019-08-13, 23:46

I recently got access to a few spare servers (all running Debian 8 Jessie) with multiple Nvidia cards. However, I'm having trouble setting up CUDA even after enabling the contrib and non-free packages and installing the required libraries. When I try to install the nvidia-cuda-toolkit package, the installer tries to run nvidia-installer-cleanup and asks whether to run nvidia-xconfig --restore-original-backup to restore the original X configuration file. Answering "Yes" causes the terminal to completely lock up while saying "No" simply terminates the installation.

I've found several threads elsewhere that mention this issue but no working solutions:

I tried using different servers but ran into the same problem every time. Anyone know how to resolve this issue?

Can I compile mfaktc on another computer and copy the binary over? Or must I have CUDA installed to run mfaktc?

hansl · 2019-08-14, 01:33

Quote:

Originally Posted by ixfd64

Can I compile mfaktc on another computer and copy the binary over? Or must I have CUDA installed to run mfaktc?

I think you still need the cuda shared libraries (same version it is compiled for) but yeah you should be able to compile on one and run on another.

I would try copying the runtime lib from the compiling machine ( "/usr/local/cuda/lib64/libcudart.so" is the path on mine) into the directory with mfaktc executable and see if that works. You might need some of the other files too like libcufft.so, etc. I'm not sure what all mfaktc uses.

VBCurtis · 2019-08-14, 01:45

In my experience, a working CUDA install is necessary in order to run any CUDA binary. You don't need to build all the tools to compile, but you do need a functioning CUDA.

chris2be8 · 2019-08-14, 16:08

See https://mersenneforum.org/showthread.php?t=23561 for details of what I had to do to get CUDA working in Devuan (that's a Debian variant without systemd so should be relevant).

Chris

ixfd64 · 2019-08-16, 22:21

Quote:

Originally Posted by chris2be8

See https://mersenneforum.org/showthread.php?t=23561 for details of what I had to do to get CUDA working in Devuan (that's a Debian variant without systemd so should be relevant).

Chris

Thanks for the information. However, my situation seems to be a bit different. From what I gather, your system already had CUDA installed whereas I'm not able to install it at all. I get an error saying that NVCC could not be found when I try to build mfaktc.

chris2be8 · 2019-08-17, 16:00

I had to start from scratch by installing CUDA before I started trying to get ECM installed.

Also search https://dev1galaxy.org/index.php (the Devuan forum) for references to CUDA (I was probably the first person to get it working on Devuan).

Chris

2019-08-13, 23:46	#1
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2³×313 Posts	problem installing CUDA on Debian I recently got access to a few spare servers (all running Debian 8 Jessie) with multiple Nvidia cards. However, I'm having trouble setting up CUDA even after enabling the contrib and non-free packages and installing the required libraries. When I try to install the nvidia-cuda-toolkit package, the installer tries to run nvidia-installer-cleanup and asks whether to run nvidia-xconfig --restore-original-backup to restore the original X configuration file. Answering "Yes" causes the terminal to completely lock up while saying "No" simply terminates the installation. I've found several threads elsewhere that mention this issue but no working solutions: http://forums.debian.net/viewtopic.php?f=6&t=123566 http://forums.debian.net/viewtopic.php?f=5&t=133946 https://bugs.debian.org/cgi-bin/bugr...cgi?bug=798732 I tried using different servers but ran into the same problem every time. Anyone know how to resolve this issue? Can I compile mfaktc on another computer and copy the binary over? Or must I have CUDA installed to run mfaktc? Last fiddled with by ixfd64 on 2019-08-13 at 23:49

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Is there any Free cuda-capable drivers for Debian?	alexvong1995	GPU Computing	22	2015-01-15 13:56
Problem in running msieve with CUDA	mohamed	Msieve	20	2013-08-01 08:27
Installing gmp under Ubuntu	R. Gerbicz	Software	3	2009-03-13 15:49
Installing GGNFS	Zeta-Flux	Factoring	33	2005-06-10 21:19
Problem running prime-net on debian (woody) dual processor	thedagit	Software	3	2002-10-19 05:57

2019-08-14, 01:45	#3
VBCurtis "Curtis" Feb 2005 Riverside, CA 2×2,927 Posts	In my experience, a working CUDA install is necessary in order to run any CUDA binary. You don't need to build all the tools to compile, but you do need a functioning CUDA.

2019-08-14, 16:08	#4
chris2be8 Sep 2009 2⁵·7·11 Posts	See https://mersenneforum.org/showthread.php?t=23561 for details of what I had to do to get CUDA working in Devuan (that's a Debian variant without systemd so should be relevant). Chris

2019-08-17, 16:00	#6
chris2be8 Sep 2009 100110100000₂ Posts	I had to start from scratch by installing CUDA before I started trying to get ECM installed. Also search https://dev1galaxy.org/index.php (the Devuan forum) for references to CUDA (I was probably the first person to get it working on Devuan). Chris