GPGPU

This page aims to compile a list of solutions on using General Purpose Graphical Processing Units for OpenFOAM (GPGPU at Wikipedia).

1 Open Source

This list includes those that have commercial support, but all have the source code licensed under an OSI approved license.

• Symscape's GPU Linear Solver Library for OpenFOAM
  • Valid versions:
  • Precision: Single-Precision only
  • Performance reports: CFD Performance Comparison Between GPU and CPU
  • Versions:
    • GPU v1.1 Linear Solver Library for OpenFOAM
    • GPU v1.0 Linear Solver Library for OpenFOAM
    • GPU v0.2 Linear Solver Library for OpenFOAM
    • GPU v0.1 Linear Solver Library for OpenFOAM

• PARALUTION
  • Valid versions:
  • Precision: unknown
  • Features: Documentation
  • Forum: Public Forums at Paralution
  • It's designed to work with CUDA.

• rocALUTION and rocSPARSE
  • These two are essentially ports of PARALUTION for AMD's ROCm solutions. More details are provided here.
  • There is no clear indication of how to use it with OpenFOAM, but it should be possible to use PARALUTION's plugin mechanism.

• cufflink-library
  • Valid versions:
  • Was fully integrated into and is supported since then.
  • Repository at Github

• RapidCFD
  • Valid versions:
  • Precision: Double-precision

2 Closed Source

The ones on this list are mostly composed of source code that is not available to their client. Usually this means that the binaries that work with the GPU act as a server (data transacted over files or similar) and are therefore not directly linked to OpenFOAM.

• SpeedIT Extreme by Vratis
  • Valid versions:
  • Precision: Single-Precision and Double-Precision

• Culises by FluiDyna
  • Valid versions:
  • Precision: Single-Precision and Double-Precision
  • Announcements: FluiDyna releases Culises 1.1 for GPU-based acceleration of OpenFOAM
  • Note: There is no entry page (18:26, 17 June 2018 (CEST)) to the public, therefore it's possible that this has been discontinued.