Root access required for CUDA?

Question

I am using GeForce 8400M GS on Ubuntu 10.04 and I am learning CUDA programming. I am writing and running few basic programs. I was using cudaMalloc, and it kept giving me an error until I ran the code as root. However, I had to run the code as root only once. After that, even if I run the code as normal user, I do not get an error on malloc. What's going on?

Answer 1

This is probably due to your GPU not being properly initialized at boot. I've come across this problem when using Ubuntu Server and other installations where an X server isn't being started automatically. Try the following to fix it:

Create a directory for a script to initialize your GPUs. I usually use /root/bin. In this directory, create a file called cudainit.sh with the following code in it (this script came from the Nvidia forums).

#!/bin/bash

/sbin/modprobe nvidia

if [ "$?" -eq 0 ]; then

    # Count the number of NVIDIA controllers found.
    N3D=`/usr/bin/lspci | grep -i NVIDIA | grep "3D controller" | wc -l`
    NVGA=`/usr/bin/lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`

    N=`expr $N3D + $NVGA - 1`
    for i in `seq 0 $N`; do
        mknod -m 666 /dev/nvidia$i c 195 $i;
    done    

    mknod -m 666 /dev/nvidiactl c 195 255

else
    exit 1
fi  

Now we need to make this script run automatically at boot. Edit /etc/rc.local to look like the following.

#!/bin/sh -e
#   
# rc.local
#   
# This script is executed at the end of each multiuser runlevel.
# Make sure that the script will "exit 0" on success or any other
# value on error.
#   
# In order to enable or disable this script just change the execution
# bits.
#   
# By default this script does nothing.

#
# Init CUDA for all users
#   
/root/bin/cudainit.sh

exit 0

Reboot your computer and try to run your CUDA program as a regular user. If I'm right about what the problem is, then it should be fixed.

Answer 2

To work with Ubuntu 14.04 I followed https://devtalk.nvidia.com/default/topic/699610/linux/334-21-driver-returns-999-on-cuinit-cuda-/ to add nvidia-uvm to etc/modules, and to add a line to a custom udev rule. Create /etc/udev/rules.d/70-nvidia-uvm.rules with this line:

KERNEL=="nvidia_uvm", RUN+="/bin/bash -c '/bin/mknod -m 666 /dev/nvidia-uvm c $(grep nvidia-uvm /proc/devices | cut -d \  -f 1) 0;'"

I don't understand why sudo modprobe nvidia-uvm works to create a proper /dev/nvidia-uvm (as does sudo cuda_program) but the /etc/modules listing requires the udev rule.