CUDA on GeForce 8600GT

Question

I have got the cuda driver, toolkit and sdk installed in Ubuntu 10.04. I'm using nVidia Geforce 8600 GT card. Official website says my card is CUDA supported. But on running the deviceQuery that comes with the cuda sdk, I'm getting the following output.

./deviceQuery Starting...

CUDA Device Query (Runtime API) version (CUDART static linking)

There is no device supporting CUDA

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 134566327, CUDA Runtime Version = 0.0, NumDevs = 0


PASSED

Press <Enter> to Quit...

Moreover, while booting, I'm getting the following error. Ubuntu is running in low-graphics mode

The following error was encountered. You may need to update your configuration to solve this.

(EE) NVIDIA: Failed to load the NVIDIA kernel module.
Please check you
(EE) NVIDIA:   system's kernel log for additional error messages.
(EE) Failed to load module "nvidia" (module-specific error, 0)
(EE) No drivers available.

But, when I restart X, it boots perfectly.

And, glxinfo gives the following output.

name of display: :1.0
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Error: couldn't find RGB GLX visual or fbconfig

Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
Xlib:  extension "GLX" missing on display ":1.0".
3 GLXFBConfigs:
   visual  x  bf lv rg d st colorbuffer ax dp st accumbuffer  ms  cav
 id dep cl sp sz l  ci b ro  r  g  b  a bf th cl  r  g  b  a ns b eat
----------------------------------------------------------------------
Segmentation fault

So, is GeForce 8600GT actually not CUDA supported, or is the problem with something else? Thanks.

Answer 1

Well. There is a good way to install the nvidia driver correctly and the avoid problems later. Here is a great howto, step-by-step, easy-to-use.

But let me correct it out, regarding the 10.04 release !
First of all (before steps), download the "dkms" pack from the bottom of the post on the linked page, and the nvidia driver from nvidia.com into your home directory.
Step1, remove the drivers. Fix the "180" to "190" or "195" , don't sure how Ubuntu calls it at the minutre.
At step 2, edit /etc/blacklist.d/blacklist.conf . Add 2 new entries to the end:
blacklist nv
blacklist nouveau

Then do a reboot, at the boot menu, select recovery mode. Go with the "root mode with networking" (or what, its at the bottom, you will be able to identify it, don't worry about the instructions. :))

When it boots, type your root password. Then type: init 3 . Login again (yay).

Now, install the driver with sudo sh ./NV* . There will be an error about "distributor provided.." don't care about it, just agree, yes yes (more, grep, fsck :)).

After it finishes, do a sudo nvidia-xconfig . THEN, do the sudo sh ./installdkms* part. After it finishes, you are done, reboot.

---

Yeah I'm aware of the howto and how its 'harder' than the "install restricted modules". However, a lot of people noticed issues , anomalies with the default driver. This way you will get the NVidia binary driver, more recent than the one Ubuntu ships, and it won't be a problem during kernel upgrades. Also, you can upgrade the driver by hand whenever you want. If you get stuck, comment, ask. (Check which part seems to be hard , check if you can find that blacklist and such before you dive in.)

---

And yeah, after this, we'll continue with the CUDA stuff. :)