Development image with the inflare library
2.1K
Development image with the Inflare RDMA streaming library installed.
Inflare is a C++ library for high-performance RDMA data streaming over IB Verbs, with optional CUDA / GPUDirect RDMA support for GPU memory streaming.
| Tag | Base | Purpose |
|---|---|---|
base | fogesque/rdma-cpp-dev-base:cpu | CPU / IB Verbs Inflare SDK |
gpunetio | fogesque/rdma-cpp-dev-base:gpu | CUDA-enabled Inflare SDK |
find_dependency()rdma-cpp-dev-basefind_package(inflare CONFIG REQUIRED)
target_link_libraries(app PRIVATE inflare::inflare)
The image sets CMAKE_PREFIX_PATH so CMake can find Inflare and its packaged dependencies.
docker run --rm -it \
--network=host \
--device=/dev/infiniband \
--cap-add=IPC_LOCK \
--ulimit memlock=-1 \
-v "$PWD:/workspace/project" \
fogesque/inflare-sdk:base
GPU builds should use the gpunetio tag and add:
--gpus all
Run inside the container:
inflare-rdma-env
This checks visible RDMA devices, verbs providers, link state, PCI/NUMA information, and CUDA/NVIDIA state when available.
The host must expose RDMA devices with --device=/dev/infiniband and allow memory locking for RDMA registration. GPU mode requires NVIDIA Container Toolkit, a compatible NVIDIA driver, CUDA-capable GPU, and GPUDirect RDMA host support where required.
Content type
Image
Digest
sha256:5a3f7dc27…
Size
3.9 GB
Last updated
3 months ago
docker pull fogesque/inflare-sdk:0.3.3-gpunetio