Gpufit

Levenberg Marquardt curve fitting in CUDA with several MRI relevant models.

This fork has been updated to support newer compilers and CUDA versions, and includes additional MRI-relevant models (Patlak, Tofts, Extended Tofts, Tissue Uptake, 2CXM, T1 FA Exponential). It also includes binaries for windows, linux and mac. The original Gpufit repository is referenced below along with the relevant citation.

Homepage for original project: github.com/gpufit/Gpufit

The manuscript describing Gpufit is now published in Scientific Reports.

Changelog

v1.4.1

• Complete Cpufit MRI model coverage — all six MRI pharmacokinetic models (Patlak, Tofts, Extended Tofts, Tissue Uptake, 2CXM, T1 FA Exponential) are now available in both GPU (Gpufit) and CPU (Cpufit) libraries with full parity.
• Fixed Patlak CPU derivative bug — the trapezoidal integral in calc_derivatives_patlak accumulated across data points instead of resetting per point, producing incorrect gradients and convergence failures.
• GPU/CPU MRI parity test — new Gpufit_Cpufit_MRI_Parity example validates that all six MRI models produce matching results between GPU and CPU.
• Updated CUDA architecture support — added Hopper (sm_90) and Blackwell (sm_100, sm_120) architectures; always emits PTX for the highest architecture to guarantee forward compatibility with future GPUs.
• Automated CI/CD releases — GitHub Actions workflows for continuous integration and automated binary releases (tag-triggered and manual dev builds) across Windows (CUDA 12.4, 12.8), Linux (CUDA 11.8, 12.5, 13.0), and macOS (CPU-only x64 and arm64).

Quick start instructions

The release artifacts provide library and wrapper payloads. To run GPU-vs-CPU performance checks, build and run the profiling examples from source (for example examples/c++/gpu_vs_cpu_profiling/Gpufit_Cpufit_Performance_Comparison.cpp).

Binary distribution

The latest Gpufit binary release can be found on the release page.

Current automated release artifacts and runtime requirements:

Artifact	Includes	Built on CI runner	Runtime requirements
windows-x64-cuda12.4	Gpufit + Cpufit	windows-2022	Windows x64; NVIDIA display driver >= 551.61 (CUDA 12.4 GA minimum)
windows-x64-cuda12.8	Gpufit + Cpufit	windows-2025	Windows x64; NVIDIA display driver >= 570.65 (CUDA 12.8 GA minimum)
linux-x64-cuda11.8	Gpufit + Cpufit	ubuntu-22.04	Linux x86_64; NVIDIA driver >= 520.61.05 (CUDA 11.8 GA minimum); userspace compatibility with Ubuntu 22.04 build baseline
linux-x64-cuda12.5	Gpufit + Cpufit	ubuntu-24.04	Linux x86_64; NVIDIA driver >= 555.42.02 (CUDA 12.5 GA minimum); userspace compatibility with Ubuntu 24.04 build baseline
linux-x64-cuda13.0	Gpufit + Cpufit	ubuntu-24.04	Linux x86_64; NVIDIA driver >= 575.51.03 (CUDA 13.0 GA minimum); userspace compatibility with Ubuntu 24.04 build baseline
macos-x64-cpu	Cpufit only	macos-15-intel	macOS x64; no NVIDIA/CUDA requirement
macos-arm64-cpu	Cpufit only	macos-14	macOS arm64; no NVIDIA/CUDA requirement

Notes:

• macOS artifacts are CPU-only and do not include the CUDA Gpufit shared library.
• CUDA driver minimums above come from NVIDIA CUDA Toolkit release notes: 11.8, 12.5, 12.8, 13.0.
• If your target machine is older/newer than the CI baseline OS, build from source for best compatibility.

Documentation

Documentation for the Gpufit library may be found online (latest documentation), and also as a PDF file in the binary distribution of Gpufit.

Building Gpufit from source code

Instructions for building Gpufit are found in the documentation: Building from source code.

Using the Gpufit binary distribution

Instructions for using the binary distribution may be found in the documentation. The binary package contains:

• Cpufit shared library in all artifacts.
• Gpufit shared library in CUDA-enabled Linux/Windows artifacts.
• C/C++ headers under include/ (Cpufit in all artifacts, Gpufit in CUDA-enabled artifacts).
• Python wrapper payloads (Cpufit and Gpufit source wrapper trees, plus wheels if generated).
• MATLAB wrapper payload when generated by the CI build.
• Java wrapper payload when generated by the CI build.
• BUILD_INFO.txt with build metadata.

MRI model support (release CUDA binaries)

Release model notes:

• Linux/Windows CUDA artifacts include these models in Gpufit.
• macOS artifacts are CPU-only (Cpufit only; no CUDA/Gpufit shared library).
• The Gpufit_Cpufit_MRI_Parity executable is included in the bin folder of the release package to verify model implementation.

Model	Included in CUDA release binaries (Linux/Windows)	CPU version (Cpufit)	macOS release support
PATLAK	Yes	Yes	CPU-only (Cpufit)
TOFTS	Yes	Yes	CPU-only (Cpufit)
TOFTS_EXTENDED	Yes	Yes	CPU-only (Cpufit)
TISSUE_UPTAKE	Yes	Yes	CPU-only (Cpufit)
TWO_COMPARTMENT_EXCHANGE (2CXM)	Yes	Yes	CPU-only (Cpufit)
T1_FA_EXPONENTIAL	Yes	Yes	CPU-only (Cpufit)

Examples

There are various examples that demonstrate the capabilities and usage of Gpufit. They can be found at the following locations:

• /examples/c++ - C++ examples for Gpufit
• /examples/c++/gpu_vs_cpu_profiling - C++ examples that use Gpufit and Cpufit
• /examples/matlab - Matlab examples for Gpufit including spline fit examples (also requires Gpuspline)
• /examples/python - Python examples for Gpufit including spline fit examples (also requires Gpuspline)
• /Cpufit/matlab/examples - Matlab examples that only uses Cpufit
• /Gpufit/java/gpufit/src/test/java/com/github/gpufit/examples - Java examples for Gpufit

Authors

Gpufit was created by Mark Bates, Adrian Przybylski, Björn Thiel, and Jan Keller-Findeisen at the Max Planck Institute for Biophysical Chemistry, in Göttingen, Germany.

How to cite Gpufit

If you use Gpufit in your research, please cite our publication describing the software. A paper describing the software was published in Scientific Reports. The open-access manuscript is available from the Scientific Reports website, here.

• Gpufit: An open-source toolkit for GPU-accelerated curve fitting
  Adrian Przybylski, Björn Thiel, Jan Keller-Findeisen, Bernd Stock, and Mark Bates
  Scientific Reports, vol. 7, 15722 (2017); doi: https://doi.org/10.1038/s41598-017-15313-9

License

MIT License

Copyright (c) 2017 Mark Bates, Adrian Przybylski, Björn Thiel, and Jan Keller-Findeisen

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.