Geek Answers Handbook

How to install an NVIDIA graphics card for display and another for computers (on Linux)?

I have the NVIDIA and CUDA display driver installed on my machine with one NVIDIA GT520 card (used for both display and calculation), but it gives me the error "start timed out and stopped" for some program. I searched for this error, they say that this error is related to OS Watchdog (CentOS 6), and my kernel takes too much time. I find one solution in which I can insert two cards, and I can use them to display and others to calculate on this link . How can I configure a specific map for display and another for calculation. I have CentOS 6.5 with the kernel 2.6.32-431.3.1.el6.x86_64.

+6
source share
1 answer

A general description of how to do this is given here . You want to use option 1, which is displayed below:

Option 1: Use two GPUs (RECOMMENDED)

If two GPUs can be available in the system, then X processing can be processed on one GPU, while CUDA tasks are executed on the other. This ensures full interactivity and interference-free X, while ensuring unhindered CUDA execution.

To accomplish this:

• The X display must be forced on to one GPU using the BusID parameter in the corresponding Device section of the xorg.conf file. In addition, any other sections of the "Device" must be deleted. For instance:

BusID "PCI:34:0:0"

GPU PCI identifiers can be determined from the lspci command or from the nvidia-smi -a command.

• CUDA processing must be forcibly transferred to another GPU, for example, using the CUDA_VISIBLE_DEVICES environment variable before running any CUDA applications. For instance:

  export CUDA_VISIBLE_DEVICES="1"

(Select a numeric option to select a GPU that is not an X-GPU)

+8
source

More articles:


All Articles