Skip to content

Gpufit source code and binary package v1.1.0 (Windows)
Compare
Choose a tag to compare
@superchromix superchromix released this 13 Apr 09:06
· 110 commits to master since this release
v1.1.0
49d07e8

Version 1.1.0 for the Gpufit library (https://github.com/gpufit/Gpufit).

Binary files compiled with Microsoft Visual Studio 2015 and CUDA toolkit 8.0.

Release notes:

In this version:

  • A new interface has been added to Gpufit (the CUDA interface), allowing better integration with other CUDA pre- and post-processing steps.
  • Adaptive scaling of the Hessian intermediate matrix is included to improve convergence speed.
  • A binding to the Java programming language has been added.
  • Improved GPU memory allocation for the user_info parameter.
  • For 64 bit architectures, the linear equation system in the Levenberg Marquardt algorithm may optionally be solved by LUP decomposition using the Nvidia cuBLAS library. The use of cuBLAS is specified at compilation, and therefore two different builds of Gpufit (with or without cuBLAS) are included in this release.
  • The maximum number of model parameters was limited to 32 in previous releases. The current release includes an option to use cuBLAS (see above), and for cuBLAS Gpufit the maximum number of model parameters is increased to 1024.

The binary package contains:

  • The Gpufit manual (PDF format)
  • Gpufit SDK (compiled for 32-bit and 64-bit Windows OS)
  • Matlab binding
  • Python binding (the pyGpufit module)
  • Java binding
  • Usage examples written in C, Matlab, Python, and Java
  • Executable application which tests Gpufit function and GPU performance
  • License information

Note that two different versions of the Gpufit build are included with this release. One which makes use of cuBLAS and runs only on 64-bit architectures, and one which does not use cuBLAS and runs on 32 or 64-bit machines.

This package was compiled on April 13, 2018.
See license statement contained in the LICENSE.txt file.

Assets 4