Overview
Examples
Screenshots
Comparisons
Applications
Download
Documentation
Tutorials
Bazaar
Status & Roadmap
FAQ
Authors & License
Forums
Funding Ultimate++
Search on this site
Search in forums












SourceForge.net Logo
Home » Developing U++ » UppHub » New package ArrayFire_demo
New package ArrayFire_demo [message #45731] Wed, 30 December 2015 18:20 Go to next message
koldo is currently offline  koldo
Messages: 3354
Registered: August 2008
Senior Veteran
ArrayFire_demo package includes simple ArrayFire benchmarks and demos.

ArrayFire is a library for fast GPU computing, supporting both Nvidia CUDA and OpenCL devices, and it is open source (BSD 3-clause).

It includes matrix algebra, algorithms, image processing, etc. classes and functions.

You can download ArrayFire sources from GitHub or the binaries from the ArrayFire page. Install is very easy, although in few days I will include instructions.

Benchmarks and demos included in package are:
- Pi: Pi number benchmark
- Matrix: NxN matrix product benchmark
- Vectorize: Different strategies to do operations between vectors
- Demo: Basic matrix algebra demos

For example in my poor Nvidia card, Pi demo works 30 times faster in GPU than in CPU.


Best regards
Iñaki
icon14.gif  Re: New package ArrayFire_demo [message #45732 is a reply to message #45731] Wed, 30 December 2015 21:18 Go to previous messageGo to next message
Didier is currently offline  Didier
Messages: 680
Registered: November 2008
Location: France
Contributor
This is very interesting,

Thanks
Re: New package ArrayFire_demo [message #45733 is a reply to message #45732] Wed, 30 December 2015 22:14 Go to previous messageGo to next message
koldo is currently offline  koldo
Messages: 3354
Registered: August 2008
Senior Veteran
An advantage of ArrayFire is that it is very easy to install and it uses internally cl/cuBLAS, cl/cuFFT, MKL, OpenMP and other basic accelerated libraries, transparently for the programmer.

If you want to try GPU computing the easy way, ArrayFire is a good option. If anybody else has interest in using directly basic CUDA or OpenCL I can upload some demo, although the difference in usability is abyssal Smile.


Best regards
Iñaki
Re: New package ArrayFire_demo [message #45800 is a reply to message #45733] Wed, 06 January 2016 21:51 Go to previous message
koldo is currently offline  koldo
Messages: 3354
Registered: August 2008
Senior Veteran
Added matrix product benchmark.

In this case the GPU (a very basic 100 cores NVidia) is up yo 600 times faster (about 120 GFlops) than the CPU with the naive algorithm.

The trick is that it uses cuBLAS library.

index.php?t=getfile&id=4917&private=0

This benchmark used a 2880 cores high end Tesla K40M and gets until 1200 GFlops, so with some money there is space for improvement Smile .
  • Attachment: capt.JPG
    (Size: 33.17KB, Downloaded 368 times)


Best regards
Iñaki
Previous Topic: New package GrabYourScreen
Next Topic: New package OpenMP_demo
Goto Forum:
  


Current Time: Thu Mar 28 11:39:14 CET 2024

Total time taken to generate the page: 0.01302 seconds