K-Tec

CUDA Massively Parallel Processing - On Your Graphics Card

  • mjc
  • mjc's Avatar Topic Author
  • Offline
  • Main Sequence
  • Main Sequence
  • Posts: 470
  • Thank you received: 20
Peeps

This post relates to emerging technologies that look promising for new data-reduction capabilities for astroimages (both amatuer and professional).

It maybe that some of you have been aware of the possibility of using a graphics card to do solve problems other than those of its originally intended purpose of rendering graphics for demanding gamesters - but I've stumbled on this over the last three days and its rocked me with its potential for the amateur (and professional) astroimager etc.

I perused the lecture / presentation videos of the Institute of Advanced Studies at Princeton and I came across the (second) link below - but please read this overview lest you think its a complete tanget:

Consider a Cray-X-MP (1984) - a supercomputer performing 800 MFLOPS (800 million floating point operations per second).

A top end Nvidia CUDA-enabled graphics card used as a co-processor (just plug it in) can now well beat this. In fact I think you can get well into the GFLOPS (10^9).

Of course supercomputers have also moved on (PetaFLOPS - 10^15) - but not at any reasonable multiple of commodity cost)

I can see this re-use of a commodity product (as a highly parallel co-processor to perform certain data-reduction tasks for astroimaging) rolling out to amteurs over the near future. A graphics card can do sophisticated interpolation, and things such as Lucy-Richardson deconvolution, denoising, PSF matching (allowing accurate photometry/astrometry) of faint undersampled images, faster then the host processor on a desktop. I have seen a few references to some of these applications on the "CUDA Zone":

www.nvidia.com/object/cuda_home.html#

The development SDKs compiler etc is free.

Nvidia Overview of CUDA Prestented to the IAS:
video.ias.edu/webfm_send/783

The left icon lets you download it if you prefer - or you can stream by clicking the "Hi-Res" or "Low-Res" link.

You can search the site for CUDA and GPU and obatin more information.

You can search for CUDA + N-Body, for example, youtube to see some wonderful simulations. Note this isn't rendering its modelling.

I wont be delving anytime soon - but I see interesting things arising from this.

Mark
13 years 2 months ago #82061

Please Log in or Create an account to join the conversation.

Time to create page: 0.037 seconds
Powered by Kunena Forum