getting started with GPUs

[Old posts from the commercial version of ArrayFire] Discussion of ArrayFire using CUDA or OpenCL.

Moderator: pavanky

getting started with GPUs

Postby melonakos » Thu Jan 06, 2011 8:32 pm

I'm certainly interested in GPU stuff for Matlab in the long run, but I have just bought a ZT machine with ATI Radeon HD5400 (nor sure if it supports these apps) and the other thing is that I really doubt that my current computation needs would be benefiting much from GPU parallelism. I do benefit from the parallel computing toolbox (using up to 4 cores on my ZT) for some of my Matlab work (stuff like fitting a handful of regression trees in parallel.)

If you have any information what kind of calculations would benefit the most from GPU acceleration, I'd be interested. There's perhgaps a chance to change my graphics card if I can convince myself if there's a considerable benefit...)


Great to hear from you. GPUs are really good at exploiting data-parallelism. If you have matrices in your code that are 128x128 (or 1x10000, etc) in size, then GPU computing will definitely give you a good speedup. Also if you have smaller problems that can be batched together with our GFOR loop, then you can get parallel benefit that way too.

Jacket works in conjunction with PCT too. You can continue to use PCT to use all 4 cores with 4 PCT workers. Then you can use Jacket commands to enable each worker to offload computation form the core to which it is assigned to the GPU.

There are lot's of examples about computations that see benefit on the GPU, here:


Jacket currently only supports NVIDIA GPUs. AMD/ATI GPU support is next up to bat.
John Melonakos (john@arrayfire.com)
User avatar
melonakos
 
Posts: 503
Joined: Tue Jun 10, 2008 9:49 am

Return to [archive-commercial] Programming & Development with ArrayFire

cron