Skip to content

CUDA & Emergent SDK setup

This page covers setting up Linux, the Emergent SDK, the NVIDIA driver, and CUDA. These are the foundational steps before any of the other dependencies.

Overview of steps:

Initial set-up before installing Linux

  1. Create a bootable Linux USB for an Ubuntu 22.04.4 ISO: ubuntu-22.04.4-desktop-amd64.iso. The default kernel is 6.5.
    • Alternatively, use balenaEtcher to make a bootable Linux USB.
  2. Ensure the following on the PC:
    • An existing video card that displays the existing OS or BIOS screen on a monitor — no need to worry about graphics driver for now.
    • At least 3 USB ports that work when you enter BIOS/installation (mouse / keyboard / Linux USB).
    • Important: disable any wired/wireless network connections during the install, otherwise it will automatically upgrade to kernel 6.8. As of writing, Emergent eSDK only supports up to kernel 6.5.
  3. Plug the bootable Linux USB into one of the active ports and boot up the PC.
  4. If the PC does not directly boot into the Linux USB, change Boot settings in BIOS:
    • Enter BIOS by pressing F2 or DEL (could be a different key on some motherboards).
    • Go into the BOOT menu and set it to boot from the Linux USB (disable all other boot options).
    • Save and reboot.
Possible issues you may encounter
  1. Keyboard/mice not working — not all USB ports may work during installation. Try plugging into different ports.
  2. Screen does not boot into the Linux USB — reformat the USB and recreate, or try a different USB stick.

Install Linux

  1. Wait until the Ubuntu desktop is visible and go step-by-step:
  2. Install Ubuntu.
  3. Choose keyboard layout, time zones, etc.
  4. Under Updates and other software, choose only Minimal Installation.
  5. Under Installation type, choose Erase Disk and Install Ubuntu.
  6. Choose default disk/partition to install Ubuntu.
  7. Set up user accounts. Do not run the software upgrade prompted near the end of install.
  8. Once installation completes and you log in, open a terminal (Ctrl + Alt + T) and check the Ubuntu and kernel versions: 
    lsb_release -a
uname -r
    Expected output: 
    Distributor ID: Ubuntu
Description:    Ubuntu 22.04.4 LTS
Release:        22.04
Codename:       jammy

    6.5.0-18-generic

Post installation — change grub configuration

This step is needed to boot into secure mode for installing NVIDIA drivers and to detect multiple GPUs.

  1. Open and edit the grub menu file (/etc/default/grub): 
    sudo gedit /etc/default/grub
  2. Set the following options (lines beginning with # are comments — keep or remove as needed): 
    # show the menu
GRUB_TIMEOUT_STYLE=menu
# set time out for the menu
GRUB_TIMEOUT=5
# pci=realloc=off is needed for adding two graphics cards — there may be other arguments in this variable; if so, keep them
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash pci=realloc=off"
  3. Save the file and update grub: 
    sudo update-grub

  4. Reboot and verify you can enter the splash screen.

    Expected output

    ubuntu splash screen

Update and install some useful software

  1. Disable automatic updates and upgrades, and only then connect to ethernet.
  2. Run an update: 
    sudo apt-get update
  3. Install required packages: 
    sudo apt-get install net-tools pkg-config make gcc libglvnd-dev git curl
  4. Set up SSH client and server: 
    sudo apt-get install openssh-client openssh-server
sudo systemctl enable ssh
sudo ufw allow ssh
  5. (Optional) Install grub-customizer to show only the required kernel if you end up with multiple kernels.
  6. Make the following directories for convenience:
    • /home/$USER/setup_files — for downloaded drivers and installation files.
    • /home/$USER/nvidia — for building dependencies.
    • /home/$USER/src — for cloned source repos.

Install Emergent SDK and dependencies

0. Ensure the Emergent quad-port NIC is plugged into the PCIe slot.

1. Download files from the Emergent FTP server:

2. Unzip and install the eSDK (assumes downloads are in /home/$USER/setup_files):

cd /home/$USER/setup_files
unzip eSDK_2_55_02_eCap_2_11_04_Ubuntu_22_04_6_5_0_x64.zip -d emergent
cd emergent
sudo ./install_eSdk.sh -i Mellanox

The script downloads and installs the Mellanox NIC driver and dependencies. On success:

Emergent eSDK & eCapture have been installed successfully.

3. Acquire a license file from Emergent Vision and copy it to /opt/mellanox:

sudo cp rivermax.lic /opt/mellanox/rivermax.lic

4. Launch eCapture (you can't acquire frames yet — NIC isn't configured):

cd /opt/EVT/eCapture
sudo ./eCapture

5. Configure network settings for the NIC. Under Settings > Network, you should see four similarly-named EVT NIC interfaces (enp97s0 and enp97s0dX).

Screenshot

network settings

In the screenshot above, enp97s0d2 shows as connected because a camera is plugged into that port. Apply the following network settings to each EVT NIC interface you want to use:

property value
MTU 9000
IPv4 Manual
Address 192.168.13.10
Netmask 255.255.255.0
Gateway 192.168.13.1

Convention: keep each interface on a different subnet — e.g. 192.168.10.10, 192.168.11.10, 192.168.12.10 for the other three interfaces.

6. Configure cameras to stream through eCapture:

  • Connect camera to the EVT NIC port using a fiber optic cable, then power on.
  • Open the eCapture app: 
    cd /opt/EVT/eCapture
sudo ./eCapture

  • If the camera IP is configured properly, it should show up under the Devices panel.

    Expected output

    devices in eCapture

  • If the camera does not show up, force an IP from Device Configuration → Force ip:

    field value
    MAC/SN the serial number on the camera
    IP address if the camera is plugged to the NIC port with IP 192.168.13.10, the camera IP should be 192.168.13.XX (XX ≠ 10)
    Subnet 255.255.255.0
    Gateway 0.0.0.0

  • Click Apply and wait a few seconds — the camera should appear under Devices.

  • Note: the above sets a temporary IP. To persist after power cycling, use Device Configuration → IP Configuration, set Persistent IP Address, then Write Configuration.
  • Test by power-cycling the camera and verify you can acquire frames.

7. Update camera firmware to 3.70. Download the firmware from Emergent FTP and move to /home/$USER/setup_files. Typical files:

  • BIN file: HB_IMX4xx_3.70.bin (may differ for your camera — but always a .bin)
  • XML file: XML_HB-7000S-C_3_70.zip

Follow the instructions in the included Readme.txt.

Install NVIDIA driver

0. Download the driver setup file. Go to NVIDIA's Linux driver archive and pick a version:

  • 535.183.06 for Linux kernel 6.5
  • 525.105.17 works well with the A6000 (or a PC with both A16 and A6000)
  • 550.90.07 works well with multiple A16

1. Run the installation file. Strongly recommended: boot into secure mode first (X driver running causes issues).

cd /home/$USER/setup_files
chmod +x NVIDIA-Linux-x86_64-535.183.06.run
sudo ./NVIDIA-Linux-x86_64-535.183.06.run

In the prompts that appear:

  • → Continue Installation
  • → Ok for removing Nouveau
  • → Yes for creating modprobe file
  • → No for 32-bit compatibility

If installation fails, check the log:

cat /var/log/nvidia-installer.log

2. Common errors and fixes:

  • gcc version mismatch. Install the expected gcc version and re-run with the CC environment variable: 
    sudo apt-get install gcc-12
sudo CC="/usr/bin/gcc-12" ./NVIDIA-Linux-x86_64-535.183.06.run
  • Missing dependencies/packages (e.g. pkg-config, libglvnd). Install them and re-run; you may need to do this multiple times since the installer only checks one at a time.

If you hit other errors and find a fix, please open a PR to add it here.

3. Verify driver installation:

nvidia-smi

This should list the GPUs accessible via the driver.

Install CUDA (depends on NVIDIA driver installation)

0. Download the installation file. We use CUDA 12.2 with driver 535.104.05. Other versions are at the CUDA toolkit archive — prefer the runfile (local) installer.

1. Run the install file:

cd /home/$USER/setup_files
sudo cuda_12.2.2_535.104.05_linux.run

Important:

  • Uncheck Driver under the CUDA Installer menu — don't reinstall the driver.
  • Check CUDA Toolkit 12.X.
  • Start installation with Install and follow prompts.
Install menu

cuda install menu

Note the path CUDA installs to (usually /usr/local/cuda).

2. Add to PATH and verify:

Edit .bashrc:

gedit /home/$USER/.bashrc

Add CUDA's path:

export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

Save, then verify:

source ~/.bashrc
nvcc --version

Enable GPU-Direct

0. Enable PCIE Above 4G Decoding and Resizable Bar. In BIOS, enable PCIE Above 4G Decoding and Resizable Bar. You may need to upgrade the BIOS if these aren't supported.

1. Load the nvidia-peer-memory module. Activate the init script start-nvidia-peermem so the system loads the module on boot:

sudo update-rc.d start-nvidia-peermem defaults
sudo /etc/init.d/start-nvidia-peermem start

Verify:

lsmod | grep nvidia_peermem

Expected:

nvidia_peermem         16384  0