Probtrackx GPU

Tool for performing probabilistic tractography on NVIDIA GPUs

Part of FSL (FMRIB Software Library)

For information about the functionality of the tool, see the PROBTRACKX User Guide

This page provides binary downloads of the GPU version of Probtrackx.


  • Download the correct probtrackx2_gpu file for your CUDA version
  • Unzip probtrackx2_gpu file
  • Copy the binary file into your $FSLDIR/bin directory
  • To execute it use: $FSLDIR/bin/probtrackx2_gpu


Ampere architecture, and CUDA 11.x binaries will be supported within FSL installer starting from FSL 6.0.6


GPU Probtrackx offers accelerations of more than 200 times using a single GPU compared to a single CPU core: 1 GPU ≈ 230 CPU cores.

a) Execution times (in logarithmic scale) and speedup (standard deviation σ is also shown) in the reconstruction of 15 tracts comparing a GPU-based with a CPU-based probabilistic tractography framework. (b) Execution times (in logarithmic scale) and speedup (and its standard deviation σ) reconstructing 27 tracts. (c) Execution times (in logarithmic scale) and speedup (and std) generating a dense connectome, taking and without taking into account the time spent for merging results from differenc CPU cores (required only in the CPU-based solution).

Coronal, sagittal and axial views comparing CPU-based and GPU-based frameworks performing probabilistic tractography and reconstructing some major white matter tracts. Each colour represents a different brain white matter tract. These paths are binarised versions of the path distributions after being thresholded at 0.5%.

Path probability map from a vertex in the Motor Cortex. The map was extracted from dense connectome matrices reconstructed with CPU-based and GPU-based probabilistic tractography applications.


If you use Probtrackx GPU in publications, please cite this paper:

Hernandez-Fernandez M., Reguly I., Jbabdi S, Giles M, Smith S., Sotiropoulos S.N. Using GPUs to accelerate computational diffusion MRI: From microstructure estimation to tractography and connectomes. NeuroImage 188 (2019): 598-615


  • Moises Hernandez Fernandez (FMRIB, University of Oxford, UK)
  • Stamatios N Sotiropoulos (FMRIB, University of Oxford, UK)
  • Mike Giles (Oxford e-Research Centre, University of Oxford, UK)
  • István Zoltán Reguly (Pázmány Péter Catholic University, Hungary)
  • Saad Jbabdi (FMRIB, University of Oxford, UK)
  • Stephen Smith (FMRIB, University of Oxford, UK)

Part of FSL (FMRIB Software Library)

License Infomation