Vulkan

If you encounter issues with Vulkan (or other software that depends on Vulkan, such as omnigibson) where the GPU device cannot be found, you can resolve it with the following steps:

Step 1: Write the ICD file

bash

cat >> /etc/vulkan/icd.d/my_nvidia_icd.json <<EOF
{
    "file_format_version" : "1.0.0",
    "ICD": {
        "library_path": "/lib/x86_64-linux-gnu/libEGL_nvidia.so.0",
        "api_version" : "1.3.277"
    }
}
EOF

In the library_path above, the library should only be libEGL_nvidia, not libGLX_nvidia. The reason is that the server does not have a desktop environment and operates in headless mode, which requires the former. For details, you can refer to the NVIDIA Documentation.

Step 2: Verification

Install necessary dependencies:

bash

apt update && apt install -y vulkan-tools libvulkan1 libsm6 libegl1

Finally, execute vulkaninfo --summary to confirm:

bash

$ export VK_ICD_FILENAMES=/etc/vulkan/icd.d/my_nvidia_icd.json     # Explicitly specify the ICD file path
$ vulkaninfo --summary
...
Devices:
========
GPU0:
        apiVersion         = 4206869 (1.3.277)
        driverVersion      = 2308620416 (0x899ac080)
        vendorID           = 0x10de
        deviceID           = 0x2684
        deviceType         = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU  <--- Here the device type is GPU, indicating it is correct. If the GPU is not recognized, it would be: PHYSICAL_DEVICE_TYPE_CPU
        deviceName         = NVIDIA GeForce RTX 4090            <--- Here you can see the correct GPU model is identified
        driverID           = DRIVER_ID_NVIDIA_PROPRIETARY
        driverName         = NVIDIA
        driverInfo         = 550.107.02
        conformanceVersion = 1.3.7.2
        deviceUUID         = 75219daa-2fe9-cefe-4994-1a598657b01f
        driverUUID         = 12adfef6-a92a-528b-8610-45df7fa0a5b8
...

Team & Collaboration

Tools & Environment

Network & Remote Connections

AI & ML

System Management

Parallel & Optimization