Installing Cuda on Linux without CUDA-Capable GPU

Accelerated Computing CUDA CUDA Setup and Installation
#1

Hi,
I’m trying to install Cuda driver and toolkit for linux 16 on x86 machine. I have no cuda capable GPU installed, but I want to be able to develop using eclipse nsight on my machine while deploying and running the code on TX1.
I’m trying to execute

sudo sh cuda_10.0.130_410.48_linux.run --kernel-source-path=/usr/src/linux-headers-4.15.0-36-generic/ --override

Also tried without the source path or override flags with no success.

I got this error from log:

-> Kernel messages:
[  135.536152] EXT4-fs (sda1): resizing filesystem from 10485248 to 15728384 blocks
[  135.896636] EXT4-fs (sda1): resized filesystem to 15728384
[  258.743405] VFIO - User Level meta-driver version: 0.3
[  258.758832] ipmi message handler version 39.2
[  258.761451] ipmi device interface
[  258.898600] nvidia: loading out-of-tree module taints kernel.
[  258.898605] nvidia: module license 'NVIDIA' taints kernel.
[  258.898606] Disabling lock debugging due to kernel taint
[  259.028640] nvidia: module verification failed: signature and/or required key missing - tainting kernel
[  259.038913] nvidia-nvlink: Nvlink Core is being initialized, major device number 240
[  259.039281] NVRM: No NVIDIA graphics adapter found!
[  259.039396] nvidia-nvlink: Unregistered the Nvlink Core, major device number 240
[  513.066556] ipmi message handler version 39.2
[  513.068272] ipmi device interface
[  513.083280] nvidia: version magic '4.15.0-34-generic SMP mod_unload ' should be '4.15.0-36-generic SMP mod_unload '
[  696.186242] ipmi message handler version 39.2
[  696.187706] ipmi device interface
[  696.204509] nvidia-nvlink: Nvlink Core is being initialized, major device number 240
[  696.204813] NVRM: No NVIDIA graphics adapter found!
[  696.204903] nvidia-nvlink: Unregistered the Nvlink Core, major device number 240
[ 1263.016055] ipmi message handler version 39.2
[ 1263.017492] ipmi device interface
[ 1263.035312] nvidia-nvlink: Nvlink Core is being initialized, major device number 240
[ 1263.035603] NVRM: No NVIDIA graphics adapter found!
[ 1263.035694] nvidia-nvlink: Unregistered the Nvlink Core, major device number 240

Do I have to have CUDA-Capable GPU for this installation?

Thanks,
Tal

#2

To install the driver, you need a CUDA capable GPU. But if you are just installing CUDA for development purposes (ie. not to run anything) then you don’t need to install the GPU driver. When you run the runfile installer as you are showing, just select “n” when prompted to install the GPU driver. Select “y” when prompted to install the CUDA toolkit.

However, this won’t install everything you need to target TX1 development. That is a cross-development setup, and you should use the JetPack to install that kind of environment.

https://docs.nvidia.com/jetson/index.html

You can find lots of useful info on the jetson TX1 forum:

https://devtalk.nvidia.com/default/board/139/jetson-embedded-systems/