You are not logged in.

#1 2009-04-03 03:28:14

Mirage
Member
Registered: 2009-03-19
Posts: 19

Getting CUDA up and running

I want to get CUDA running on my x64 archlinux, I've looked around several places with no real answers to how exactly I can get this going though, anyone mind pointing me in the correct direction for it?

Offline

#2 2009-04-03 04:26:39

iBertus
Member
From: Greenville, NC
Registered: 2004-11-04
Posts: 2,228

Re: Getting CUDA up and running

You can find the packages in AUR. Bascially, you'll need the official nvidia driver installed and the before mentioned packages installed. Then you can build the demos and play around. I should really give this another shot myself since I haven't tried CUDA since switching to a pair of GTX280s from my old 8800GTs.

Offline

#3 2009-04-03 18:13:53

Mirage
Member
Registered: 2009-03-19
Posts: 19

Re: Getting CUDA up and running

Laptops got a 9650GT @ 1 GB ram, it works well under server 08 but yah.

Trying to figure out how exactly to get them out of AUR, I know you need yaourt but i've been unable to get it, or find it with it..

Offline

#4 2009-04-03 18:43:12

SiC
Member
From: Liverpool, England
Registered: 2008-01-10
Posts: 430

Re: Getting CUDA up and running

You don't need yaourt but it helps, go to the archlinux.fr website and details for getting it are there.

You can download the PKGBUILD from the AUR website and then put it in a build directory and run makepkg.  This will build it, and tell you if you need any dependencies before it does so.  You can then just run pacman -U <pkgname>.pkg.tar.gz and install it.

Easy smile

Offline

#5 2009-04-03 18:45:02

Mirage
Member
Registered: 2009-03-19
Posts: 19

Re: Getting CUDA up and running

What a pain tongue 

I wonder why CUDA isn't in the community repo's or extra?

Offline

#6 2009-04-03 18:50:30

skottish
Forum Fellow
From: Here
Registered: 2006-06-16
Posts: 7,942

Re: Getting CUDA up and running

Mirage wrote:

What a pain tongue 

I wonder why CUDA isn't in the community repo's or extra?

Register at the AUR page and vote for the package. It currently has 49 votes which makes it eligible to be in community, provided that there aren't licensing issues. Every vote is helpful if you want it in the repos.

Offline

#7 2009-04-03 20:24:27

iBertus
Member
From: Greenville, NC
Registered: 2004-11-04
Posts: 2,228

Re: Getting CUDA up and running

It's really not that bad to install from AUR. You just need to download the files to some place on your disk and run makepkg from that directory. Most time the package will build and you can then install it with pacman.

Offline

#8 2009-04-03 23:34:46

Mirage
Member
Registered: 2009-03-19
Posts: 19

Re: Getting CUDA up and running

Well I got cuda going, now doing Pyrit (got that going easily)

Now i'm on cpyrit_cuda and I had to copy stub-64.h to stub-32.h (to get it to quit complaining about it not existing)

Now in trying to start pyrit benchmark i get

print >>.sys.stderr, "Failed to load CUDA-core (%s: %s)." % (e.message, _cpyrit_cuda.getLastError())
Failed to load Cuda-core (Failed to launch CUDA-kernel: CUDA_ERROR_LAUNCH_FAILED)

Last edited by Mirage (2009-04-04 00:04:35)

Offline

#9 2009-06-18 06:55:45

lwpro
Member
Registered: 2009-03-17
Posts: 11

Re: Getting CUDA up and running

Is it in the AUR? and the 2.2 version?

Offline

#10 2009-06-18 09:46:05

tomk
Forum Fellow
From: Ireland
Registered: 2004-07-21
Posts: 9,839

Re: Getting CUDA up and running

Did you look?

Offline

#11 2010-01-28 16:50:48

B15HOP
Member
From: Australia
Registered: 2005-02-10
Posts: 138

Re: Getting CUDA up and running

I know this is old an old thread but there are a few things to mention here:

Cuda doesn't like the newer GCC 4.4 because it's really targeted towards GCC 3.4. There are more checks that help make better code but fail with the current Arch GCC 4.4. AUR package is: http://aur.archlinux.org/packages.php?ID=23597 which is missing this information. The information here on the nvidia forums proves this: http://forums.nvidia.com/index.php?showtopic=107881 .
A way to solve is also mentioned there by adding some flags to the nvcc Makefile:

NVCCFLAGS := --compiler-options -fno-inline

The other thing that isn't mentioned. The AUR package puts nvcc in /usr/bin/nvcc instead of /opt/....... which means some tweaking to allow portability.
Finally, there might also be a missing dependency "pacman -Sy patch" to allow AUR to compile and download.

It's about time CUDA is allowed in the community repo because it's popular enough. Maybe a new thread could be started to push this...


"The ecological crisis is a moral issue."

Offline

Board footer

Powered by FluxBB