Department of Computer Science, UNC Chapel Hill

GPUFFTW: Download Version 1.0


We ask that you fill out a request form to identify yourself and indicate your interest in GPUFFTW. The request form will automatically direct you to our source package once you fill in the details and submit the request.

This release contains documentation, the source code for the GPUFFTW library, and an example program. GPUFFTW can be compiled on the Win32/Linux platforms and requires a programmable NVIDIA graphics card. Visual Studio .NET solution files for use on windows platform are included.

GPUFFTW is free for noncommercial use only. Click here for the terms of this distribution.

Click on a version link and fill the request form to download the distribution.

Note: GPUFFTW does run correctly on Windows XP and 8800 GTX using the latest NVIDIA drivers 158.19. It also works on Windows Vista and earlier NVIDIA GPUs and drivers

Version History

Version 1.0: GPUFFTW1.0

Requires a programmable NVIDIA Graphics Card (NVIDIA GeForceFX series, GeForce 6 series, GeForce 7 and GeForce 8 series), latest NVIDIA drivers (7772 or higher), and GLUT libraries .

Note: We will support ATI GPUs in future releases. NVIDIA linux drivers may have some performance overhead. Performance will also vary with the GPU used, and for reasonable performance, we recommend a PC with AGP8X or PCI-Express NVIDIA GeForce 7800 GT or faster cards.

Version 2.0: GPUFFTW2.0

Simon Potvin and Jerome Genest of COPL Universite Laval added two additional functions to perform multiple power-of-two real/complex 1-D FFTs on GPUs. The new version includes a fix to extend GPUFFTW to 64-bit Windows platform. Simon and Jerome also added an update on Mathworks website to enable GPUFFTW use in Matlab freely for non-commercial use.

Author: Naga K. Govindaraju

Maintained by:

Copyright 2006

Last Modification: May 22, 2006

©2003 Department of Computer Science, UNC Chapel Hilll