Installing fVDB

fVDB depends on PyTorch, and requires a CUDA-capable GPU. Below are the supported software and hardware configurations.

Software Requirements

The following is a compatibility matrix of the versions of software compatible with each minor release of fVDB. These are the versions of software fVDB was built and tested against that are officially supported:

fVDB Version	Operating System	PyTorch Version	Python Version	CUDA Version	Vulkan Version (only for visualization)
0.4	Linux Only	2.10.0	3.10 - 3.13	12.8, 13.0	1.3.275.0
0.3	Linux Only	2.8.0	3.10 - 3.13	12.8	1.3.275.0

Driver and Hardware Requirements

The following table specifies the minimum NVIDIA driver versions and GPU architectures needed to run fVDB:

Operating System	Driver Version	GPU Architecture	Compute Capability
Linux Only	550.0 or later	Ampere or later	8.0 or greater

Installation from conda-forge

To install fvdb-core in a conda environment, run the following command to install the latest released version of fvdb-core from conda-forge:

conda install --channel conda-forge fvdb-core

Installation from pre-built wheels

To install fvdb-core using pip, run the appropriate pip install command for your Pytorch/CUDA versions. These commands will install the correct version of fvdb-core if it is not already installed.

PyTorch 2.10.0 + CUDA 13.0

pip install fvdb-core==0.4.2+pt210.cu130 --extra-index-url="https://d36m13axqqhiit.cloudfront.net/simple" torch==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu130

PyTorch 2.10.0 + CUDA 12.8

pip install fvdb-core==0.4.2+pt210.cu128 --extra-index-url="https://d36m13axqqhiit.cloudfront.net/simple" torch==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu128

Note

Visualization and viewer features additionally require the nanovdb_editor Python package. Install it using the optional ‘viewer’ dependencies, by adding [viewer] to the fvdb-core package name, for example: pip install fvdb-core[viewer]==….

Installation from nightly builds

Nightly wheels are built from the latest main branch and published daily. Each nightly version is anchored to the next upcoming release recorded in pyproject.toml (currently 0.5.0) and carries a date stamp plus PyTorch/CUDA build identifiers, for example 0.5.0.dev20260428+pt210.cu130.

Under PEP 440 ordering, each nightly sorts between the in-development version on main and the corresponding final release, so passing --pre together with the nightly index URL will track the latest nightly until that release ships, then prefer the final release once it is tagged.

Latest nightly (any supported PyTorch/CUDA build)

pip install --pre fvdb-core --extra-index-url="https://d36m13axqqhiit.cloudfront.net/simple-nightly" torch==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu130

Note

The nightly index hosts wheels for every supported PyTorch/CUDA combination in a single project listing. Without an explicit local-version pin, pip selects the highest local version, which today is the CUDA 13.0 build. To target a different build (for example, CUDA 12.8) or pin a specific date for reproducibility, use one of the explicit commands below.

PyTorch 2.10.0 + CUDA 13.0

pip install fvdb-core==0.5.0.dev20260428+pt210.cu130 --extra-index-url="https://d36m13axqqhiit.cloudfront.net/simple-nightly" torch==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu130

PyTorch 2.10.0 + CUDA 12.8

pip install fvdb-core==0.5.0.dev20260428+pt210.cu128 --extra-index-url="https://d36m13axqqhiit.cloudfront.net/simple-nightly" torch==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu128

To list all available nightly versions:

pip index versions fvdb-core --index-url="https://d36m13axqqhiit.cloudfront.net/simple-nightly" --pre

Note

Replace 20260428 with the desired nightly date. Nightly builds are retained for 30 days.

Installation from source

Note

For more complete instructions including setting up a build environment and obtaining the necessary dependencies, see the fVDB README.

Clone the fvdb-core repository.

git clone git@github.com:openvdb/fvdb-core.git

Next build and install the fVDB library

pushd fvdb-core
./build.sh install verbose
popd