Drv program cannot exec without CUDA-capable GPU?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I’m now trying to execute sample projects on CentOS5 PC without CUDA-capable GPU.

I confirmed that some projects - deviceQuery, BlackScholes, matrixMul, simpleTexture - can be executed on this PC.

But other some projects - matrixMulDrv, simpleTextureDrv - cannot be executed.

At first, console shows below:

./matrixMulDrv: error while loading shared libraries: libcuda.so.1: cannot open shared object file: No such file or directory

so I get libcuda.so.1.177.67 from grahicsdriver and link it.

Then, console shows below:

NVIDIA: could not open the device file /dev/nvidiactl (No such file or directory).

 cutil error: no devices supporting CUDA

so I have no idea to solve this program.

This problem may be occured in executing DriverAPI programs.

Are there antidotes in documents or so ?

thanks.

2

Try starting X.

3

I got the error message below again.

NVIDIA: could not open the device file /dev/nvidiactl (No such file or directory).

cutil error: no devices supporting CUDA

In root user, I got other message below.

FATAL: Module nvidia not found.

NVIDIA: failed to load the NVIDIA kernel module.

cutil error: no devices supporting CUDA

Can I force to load some kernel module ?

4

If no nvidia module is found, then your driver installation is either missing or incomplete. I’d suggest (re)installing the driver again, then attempting to start X.

Generating and attaching an nvidia-bug-report.log would be helpful.

5

Umm… I’m not using CUDA-capable GPU, and I want to execute in emulation mode. Umm…

6

Please see the documentation’s discussion of HOSTEMU (GPU emulation) mode, as well as other posts on this forum.

7

OK, I was misled into thinking that I cannot install NVIDA Graphics Driver to Linux without nvidia-GPU.

But, I got error messages below while installing driver:

The version of gcc doesnot differ.(I use CentOS5.1 i386)

I cannot understand this problem, but my kernel doesnot use such modules.

I have no plan to solve this problem.

8

This may simply mean that CUDA is only ablee to emulate programs with the Runtime API, not the Driver one… that’s quite sad for those of us who cannot work with the high level API (most tools are actually only intended to be used with the high level API, not the driver one…)

9

If you lack a CUDA capable GPU, then you can’t run CUDA enabled binaries. Please see the documentation’s discussion of HOSTEMU (and/or search for HOSTEMU on this forum).

10

On my machine, I solved that by uninstalling all the apt packages with nvidia binaries (nvidia-glx, common, legacy, etc.). The only nvidia* package left should be nvidia-settings (if you use it).

EDIT: ooops, 2-years late answer… :P