GPU execution is a technique for high-performance machine learning, financial, image processing and other data-parallel numerical programming. The following options are available for executing F# on the GPU.
If you would like to list an option here, please submit a pull request byediting this page.
Option 1 - Use Alea GPU V3, for F#-enabled CUDA programming
Alea GPU is a GPU programming toolchain supporting
- Freecommunity edition for GeForce and Quadro GPUs with aCreative Commons license
- Simplified GPU programming with GPU parallel-for and parallel aggregate
- Automatic memory management to move data to and from the GPU
- GPU scripting for rapid prototyping in the interactive console
- GPU JIT compilation
- Seamless integration of NVIDIA GPU libraries such as cuRand, cuBlas, cuDNN etc.
- Debugging F# GPU code in Visual Studio with Visual NSight
- Profiling F# GPU code in Visual Studio with NVProf
- Largegallery of samples
- Also supporting C#
Alea GPU is a complete solution to develop CUDA accelerated GPU applications on .NET. It is a full compiler based on F# and LLVM to generate highly optimized GPU code.Alea GPU performs at the same level as CUDA C/C++ or Fortran code.
Alea TK is a new open source machine learning library for .NET based on Alea GPU, which shows how to use Alea GPU in larger projects.
Option 2 - Use StatFactory’s FCore library, a GPU-enabled F# maths/stats library
Option 3 - Use FSCL, an open-source F#-to-OpenCL compiler
- Seethe FSCL project blog. FSCL Compiler is a source-to-source compiler that translates quoted F# function calls and other contructs into valid C99 OpenCL kernel sources, enabling programming OpenCL-enabled parallel devices from within F#. Contributions welcome!
Option 4 - Use GpuLINQ, an open source F#/C# LINQ-to-OpenCL compiler
Option 5 - Use Brahma.FSharp, an open source F# tool for OpenCL programming
