Cuda on Matlab: Functions

Accelerated Computing CUDA CUDA Programming and Performance
1

Hey there,

I’m currently trying to figure out how to use cuda in my matlab-code! I just wonder if there are only those functions listed on the following page (under “Using Built-in Functions on GPUArray”), which I can use in my calculations which should run on the GPU:

http://www.mathworks.com/help/toolbox/distcomp/bsic4fr-1.html#bsloua3-1

?? If that’s right what I say, how to compute integrations and derivations of my data on my GPU?
Thanks so far!

2

We’ve heard similar things from everyone who’s looked at using the Parallel Computing Toolbox for GPU computation. Jacket has over 10X the functionality of PCT. You can easily compute integrations and derivatives with Jacket. Checkout the supported Jacket function list, here: http://wiki.accelereyes.com/wiki/index.php/Jacket_Function_List

Cheers!

3

Thanks so far: So there’s really no derivation and integration in the default matlab toolbox? Oo

4

The matlab toolbox is in it’s first version. I am sure 2011a will have more functionality added, just like Jacket has grown a lot through the years.

Short term, you’re better off using Jacket for anything complicated without writing CUDA code yourself. Long term I have no idea.

5

Ohh okay, do you know roughly when 2011 will be released?

@melonakos: Can find any informations about the pricing… Oo Am I blind? Please post me a link to price table :)

edit: is there any student/academic version or version to use when doing a bachelor thesis?

6

Just to be clear, there is a lot more to making GPU stuff go fast in MATLAB than just providing GPU-enabled functions. From all benchmarks we’ve seen on more than one-line of GPU MATLAB code, Jacket has much better performance. Learn more here: http://www.accelereyes.com/products/compare

Pricing is here: http://www.accelereyes.com/products/licensing

7

Just like normal: 2011a around april I think, 2011b around october/november? But long term for me is 2 - 3 years, I don’t expect Mathworks to be able to catch up before that timeframe.

8

Okay thanks so far,

and thanks to riedijk too :)

9

You can always bypass the wrappers and port cuda directly a la mex functions. It’s not as bad as it sounds, really. And you get all the conrtol and speedup as well.