Can I use Parallel processing toolbox directly on GPU (Geforce GT 520MX). How.

I would like to use image clustering with GPU. Please provide me an example on it. OpenCl is a platform but its a platform not a tool. Please clearify if any other method is availbe directly to utilize GPU.

Answer 1

[Jason Ross]

To use the GPU with MATLAB, only CUDA is supported and you need the Parallel Computing toolbox.

You must also have a GPU with compute capability 1.3 or higher ( http://developer.nvidia.com/cuda/cuda-gpus). It looks like your card has a compute capability of 2.1 so you should be OK.

I would offer a word of caution regarding the performance of the GPU -- according to the specs for the GPU ( http://www.geforce.com/hardware/notebook-gpus/geforce-gt-520mx/specifications), it has only 48 CUDA cores, and the amount of memory is not specified. The performance of this card isn't likely to give much of a performance boost -- as a comparison the GTX 680 has 1536 cores and 2 GB RAM on the card, and the Tesla C2075 has 448 cores / 6 GB RAM.

For examples, there are a number here: http://www.mathworks.com/help/distcomp/examples/index.html#gpu

[Kapil Nagwanshi]

Thanx Jason, Could u be able to tell me the procedure for linking CUDA core with MATLAB, I am using MATLAB 2010a. Exactly I saw the cores are 48 but the memory sharing has not been cleared I asked the same for, one of my professrs he told me 4*12 arbitration, but still doubt is not cleared. Anyways I am only interested to use GPU with matlab later on I will be applying it for more powerful gpus.

My Another question is how can I use the capabilities of ATI HD 5570 1gb gpu with the same extension as I pointed in previous question. I am having both kind of machine.

Thank you so much once again Jason.

KAPIL

[Walter Roberson]

Mathwork's Parallel Processing Toolbox does not support any ATI devices at this time.

There is a company named Accelereyes that has a MATLAB parallel processing toolbox that supports OpenCL devices including ATI.

[Kapil Nagwanshi]

Thanx Walter.

[Jill Reese]

Kapil,

MATLAB Parallel Computing Toolbox introduced GPU support in R2010b, so you would need to upgrade to at least that release in order to use your GPU with MATLAB.

Best,

Jill

[Kapil Nagwanshi]

THANX JILL, I WILL TRY TO EVALUATE IT AS SOON AS I REACH TO LAB..KINDLY SEND ME ANY LINK FOR EXAMPLES REGARDING THIS.

[Jason Ross]

To use MATLAB with the GPU, install the proper driver from nVidia's website -- make sure you get the CUDA enabled driver. MATLAB will find the device and you can start using it.

There have been a number of new GPU-related features added each release since 2010b, so I would recommend you look into getting the most recent release that you can. The link I provided has the relevant examples in it.