# Install TIGRE runs on NVIDIA GPUs. It is available for Python and MATLAB; both versions share the same CUDA code and the same set of algorithms. :::{important} TIGRE is **not** on PyPI. `pip install tigre` installs an unrelated package with the same name, and the old `pytigre` package on PyPI is an outdated v0.1 release. ::: ## Choose your route | You want to… | Use | Needs CUDA toolkit and compiler? | |---|---|---| | Use TIGRE from Python on Linux | [Prebuilt wheel](python.md#prebuilt-wheels-linux) | No, only an NVIDIA driver | | Use TIGRE from Python on Windows, or a Python version without a wheel | [Install from GitHub with pip](python.md#install-from-source) | Yes | | Use the conda ecosystem | [CCPi conda package](python.md#conda-unofficial) (unofficial) | No | | Use TIGRE from MATLAB | [Compile with `Compile.m`](matlab.md) | Yes | | Change TIGRE's code | [Install from a clone](python.md#from-a-local-clone) | Yes | ## Requirements at a glance - An NVIDIA GPU. When you compile TIGRE yourself, it is built for every GPU architecture your CUDA toolkit supports, so any GPU that your CUDA version supports will work. - Python 3.11 to 3.14, or MATLAB R2016b or newer. - To compile: the CUDA toolkit and a C++ compiler (MSVC on Windows, gcc on Linux). ```{toctree} :maxdepth: 2 python matlab ```