fogesque/rdma-cpp-dev-base repository overview

⁠fogesque/rdma-cpp-dev-base

C++ development base image for RDMA / IB Verbs projects.

This image is the common build environment used by Inflare. It intentionally does not depend on NVIDIA DOCA images. Instead, it uses Ubuntu or NVIDIA CUDA base images plus the standard Linux RDMA userspace stack.

⁠Tags

⁠Included Tooling

• GCC 14 / G++ 14
• CMake
• Ninja
• vcpkg
• Git, curl, pkg-config, archive tools
• rdma-core
• libibverbs-dev
• ibverbs-providers
• librdmacm-dev
• ibverbs-utils
• rdmacm-utils
• infiniband-diags
• perftest
• mstflint
• iproute2, ethtool, tcpdump
• pciutils, numactl, hwloc
• gdb, strace, lsof, procps, kmod

⁠RDMA Inspection Helper

The image includes:

inflare-rdma-env

It prints visible /dev/infiniband nodes, RDMA links/resources, verbs devices, IB status, PCI devices, NUMA topology, CUDA/NVIDIA state when available, and relevant loaded kernel modules.

⁠Example

docker run --rm -it \
    --network=host \
    --device=/dev/infiniband \
    --cap-add=IPC_LOCK \
    --ulimit memlock=-1 \
    fogesque/rdma-cpp-dev-base:cpu

For the GPU tag, also pass:

--gpus all

⁠Host Requirements

The container does not install host drivers. The host must provide working RDMA kernel drivers, matching provider support, and visible /dev/infiniband devices. GPU use also requires NVIDIA Container Toolkit and a compatible NVIDIA driver. GPUDirect RDMA setups may require nvidia-peermem on the host.