News

The content has been bookmarked!

There was an error bookmarking this content! Please retry.

GPU.NET 2.0 Brings HPC to Linux and Mac

Posted by Abel Avram on Jul 23, 2011

Sections: Operations & Infrastructure,; Architecture & Design,; Development
Topics: .NET ,; Linux ,; Windows ,; Programming ,; Operating Systems ,; Performance & Scalability ,; HPC ,; MacOS ,; GPU

GPU.NET 2.0 supports Mono, enabling building and deploying computational intensive applications for Linux and Mac OS X along the already supported Windows.

GPU.NET is a managed solution for writing computational intensive .NET applications that run on the GPU. The platform has its own compiler and runtime in order to avoid intermediary libraries that might slow it down and to generate cross platform binaries that can run on multiple systems. Currently only C# and F# are supported, but VB.NET support is to be added. GPU.NET runs on CUDA 4.0 NVIDIA graphics cards, with support for AMD devices being under development. GPU.NET is integrated with Visual Studio 2010. The article Targeting the GPU with GPU.NET contains more information on how GPU.NET works and about its plug-in architecture.

The main addition to version 2.0 is support for Mono which enables deploying GPU applications on Linux and Mac OS X. Other enhancements are on-device random number generation and double precision support.

TidePowerd, the company that is developing GPU.NET, plans to add support for two NVIDIA math libraries, CUBLAS (PDF) and CUFFT (PDF). CUBLAS is an implementation of Basic Linear Algebra Subprograms (BLAS) on top of CUDA, while CUFFT is a Fast Fourier Transform (FFT) library built on top of the same platform.

A somewhat similar solution is the open source WebCL, a binding allowing JavaScript developers to write web applications that make use of the GPU for intensive calculations, including those needing parallel hardware resources.