Installation

RiemannEngine ships as a prebuilt static library + headers + CMake package. Your program also links against LibTorch and the CUDA toolkit, and both must match the variant you install.

  1. Choose a variant
  2. Release assets
  3. Windows
    1. 1. Install RiemannEngine
    2. 2. Install the prerequisites
    3. 3. Runtime DLLs
  4. Ubuntu
    1. 1. Install RiemannEngine
    2. 2. Install the prerequisites
    3. 3. Install LibTorch
  5. Docker
  6. Verify the install

Choose a variant

Each release is built once per CUDA version. Pick the row that matches your OS and the CUDA version you have, or want, installed.

Variant OS CUDA toolkit cuDNN LibTorch Compiler
cu132 Windows 10/11 x64 13.2 9.x for CUDA 13 2.15.0+cu132 VS 2022 (MSVC 14.4x)
cu134 Windows 10/11 x64 13.4 9.x for CUDA 13 2.15.0+cu134 VS 2022 (MSVC 14.4x)
cu124 Ubuntu 22.04 x64 12.4 9.x for CUDA 12 2.5.1+cu124 (CXX11 ABI) GCC 11
cu132 Ubuntu 24.04 x64 13.2 9.x for CUDA 13 2.15.0+cu132 GCC 13
cu134 Ubuntu 24.04 x64 13.4 9.x for CUDA 13 2.15.0+cu134 GCC 13

GPU architecture. Release binaries contain kernels for compute capability 8.6 (RTX 30-series / A-series Ampere). Other GPUs need a source build with CMAKE_CUDA_ARCHITECTURES set to your SM number. CUDA 13 cannot target GPUs older than Turing (SM 7.5); use the cu124 variant or a CUDA 12 source build for those.

Release assets

Download from the Releases page.

Asset Contents Installs / extracts to
RiemannEngine-<ver>-cu<X>-x64.msi Windows installer C:\Program Files\RiemannEngine\<ver>_cu<X>\
RiemannEngine-<ver>-cu<X>-win-x64.zip Windows SDK, no installer RiemannEngine\<ver>_cu<X>\
libriemann-cpp-dev_<ver>_cu<X>_ubuntu<U>_amd64.deb Ubuntu package /usr/local/RiemannEngine/<ver>_cu<X>/
RiemannEngine-<ver>-cu<X>-ubuntu<U>-x64.tar.gz Ubuntu SDK, no package manager RiemannEngine/<ver>_cu<X>/
SHA256SUMS.txt Checksums for every asset  

Every install has the same layout, and variants install side by side:

RiemannEngine/<ver>_cu<X>/
├── include/riemann/              # #include <riemann.hpp>
│   ├── riemann.hpp               # umbrella header
│   ├── engine/ octree/ splat/    # canonical headers
│   └── layers/ dynamic/ normal/  # forwarding headers
└── lib/
    ├── riemann_cpp.lib           # Windows (libriemann_cpp.a on Linux)
    ├── libiomp5md.lib            # Windows only: OpenMP import lib matching LibTorch
    └── cmake/RiemannEngine/      # RiemannEngineConfig.cmake + targets

Verify a download:

sha256sum -c SHA256SUMS.txt --ignore-missing          # Linux

Get-FileHash .\RiemannEngine-1.0.0-cu134-x64.msi       # Windows, compare by eye

Windows

1. Install RiemannEngine 

msiexec /i RiemannEngine-1.0.0-cu134-x64.msi

Or extract the .zip anywhere, for example C:\, to get C:\RiemannEngine\1.0.0_cu134.

2. Install the prerequisites

Requirement How
NVIDIA driver Latest GeForce/RTX/Studio driver. nvidia-smi must report a CUDA version ≥ your variant.
CUDA Toolkit 13.2 / 13.4 CUDA Toolkit archive. The installer sets CUDA_PATH.
cuDNN 9 cuDNN downloads → Windows / x86_64 / CUDA 13
Visual Studio 2022 Desktop development with C++ workload
CMake ≥ 3.24, Ninja winget install Kitware.CMake Ninja-build.Ninja
LibTorch 2.15.0 See below 
$cu = "cu134"   # or cu132
curl.exe -L -o libtorch.zip "https://download.pytorch.org/libtorch/$cu/libtorch-win-shared-with-deps-2.15.0%2B$cu.zip"
tar -xf libtorch.zip -C C:\
Rename-Item C:\libtorch "C:\libtorch_$cu"

Use the Release LibTorch build. The Debug build is ABI-incompatible with the Release riemann_cpp.lib.

3. Runtime DLLs

Programs need LibTorch’s DLLs (torch_cuda.dll, c10.dll, and others) at run time. Either add C:\libtorch_cu134\lib to PATH, or copy the DLLs next to your .exe. The quick-start CMake file does the copying for you.

Ubuntu

1. Install RiemannEngine 

sudo apt install ./libriemann-cpp-dev_1.0.0_cu134_ubuntu24.04_amd64.deb

Or, without a package manager:

sudo tar -xzf RiemannEngine-1.0.0-cu134-ubuntu24.04-x64.tar.gz -C /usr/local

Uninstall with sudo apt remove libriemann-cpp-dev-1.0.0-cu134.

2. Install the prerequisites 

# NVIDIA driver
sudo ubuntu-drivers install && sudo reboot

# CUDA + cuDNN from NVIDIA's repository (use ubuntu2204 on 22.04)
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update
sudo apt install cuda-toolkit-13-4 cudnn9-cuda-13     # cu134
# sudo apt install cuda-toolkit-13-2 cudnn9-cuda-13   # cu132
# sudo apt install cuda-toolkit-12-4 cudnn9-cuda-12   # cu124 (22.04)

# Build tools
sudo apt install build-essential cmake ninja-build unzip wget libgomp1

Don’t install Ubuntu’s own nvidia-cuda-toolkit package. It is CUDA 12.0 and conflicts with NVIDIA’s packages.

3. Install LibTorch 

cu=cu134   # or cu132
wget -O libtorch.zip "https://download.pytorch.org/libtorch/$cu/libtorch-shared-with-deps-2.15.0%2B$cu.zip"
# cu124 (Ubuntu 22.04):
# wget -O libtorch.zip "https://download.pytorch.org/libtorch/cu124/libtorch-cxx11-abi-shared-with-deps-2.5.1%2Bcu124.zip"
sudo unzip -q libtorch.zip -d /opt          # -> /opt/libtorch

Pass the LibTorch folder to CMake with -DCMAKE_PREFIX_PATH=/opt/libtorch; your own find_package(Torch) call needs it. (RiemannEngine’s package also falls back to /opt/libtorch when it looks for Torch itself.)

Docker

The acaicia/riemannengine images contain CUDA, cuDNN, LibTorch, CMake, Ninja and RiemannEngine, already built and installed to /usr/local.

Tag Ubuntu CUDA LibTorch
cu134, latest, <ver>-cu134 24.04 13.4.3 2.15.0
cu132, <ver>-cu132 24.04 13.2.1 2.15.0
cu124, <ver>-cu124 22.04 12.4.0 2.5.1

Host requirements:

  • Linux: Docker Engine plus the NVIDIA Container Toolkit. Run sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker.
  • Windows: Docker Desktop with the WSL 2 backend and a current NVIDIA driver.
docker run --gpus all --rm -it -v "$(pwd):/workspace" acaicia/riemannengine:cu134
# inside the container:
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH=/opt/libtorch
cmake --build build
Path in the image Contents
/usr/local/include/riemann Headers
/usr/local/lib/libriemann_cpp.a Static library
/usr/local/lib/cmake/RiemannEngine CMake package
/opt/libtorch LibTorch
/workspace Working directory

See the Docker guide for host setup, file ownership, Compose, VS Code Dev Containers, slim deployment images, supported GPUs and troubleshooting.

Verify the install

Build and run the quick-start program. It prints the version and the device, and runs a small network on the GPU.