Notice for NVIDIA RDMA

NVIDIA RDMA is only suppported on Linux.

NVIDIA RDMA samples require a NVIDIA GPU that suppports RDMA.

The NVIDIA RDMA samples allocate memory on the GPU and announce this memory using NvidiaRdmaMemory.

See the following files in the eGrabber sample programs:

●

cpp/egrabber/samples/503-grabn-cuda-rdma-process.*

●

cpp/nvidia/egrabber-cuda with the command line argument cudaRDMA

Installation instructions:

●

NVIDIA CUDA drivers:

□

Follow the installation instructions from: https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_network

●

NVIDIA driver sources:

□

These are needed to produce the Module.symvers file associated with the installed nvidia driver. This file will be required to install the eGrabber package

□

Select the appropriate driver from https://www.nvidia.com/download/index.aspx?lang=en-us

□

Make sure to download the version that matches the nvidia-<version> already installed in /usr/src/

□

Extract the archive with -x option

□

Change to directory kernel in the extracted archive

□

Run make module

□

The file Module.symvers should have been generated

●

eGrabber package

□

Extract the egrabber-linux-x86_64 archive

□

Install the package with the following command: sudo NVIDIA_KERNEL_PATH=<dir path containing Module.symvers> ./install.sh

The line Enabling NVIDIA RDMA build should appear during the installation of the eGrabber package.

A successful build can be confirmed if the command lsmod | grep coaxlink (or lsmod | grep grablink) indicates that coaxlink (or grablink) module depends on nvidia.