[Eeglablist] CUDAICA: GPU Optimization of Infomax-ICA EEG Analysis

[Tarik S Bel-Bahar]

Thanks very much for your kind and clear responses!

On Wed, Oct 17, 2012 at 6:15 AM, Federico Raimondo <fraimondo at dc.uba.ar>wrote:

> Greetings,
>
>
> 1. How much minimal RAM is needed to run CUDAICA ?
>
>
> It all depends on the size of the dataset. You'll need at least NCHANNELS
> * NSAMPLES * 8 bytes of both CPU RAM and GPU RAM memory for the dataset.
> There are also more data structures involved, so keep in mind that if you
> have a big dataset, it might not fit into the GPU.
>
>
> 2. Whether it runs on Mac OS 10.6 ?
>
>
> It should, but I don't have such version to test on it.
>
>
> 3. Whether any of the needed downloads (such as Xcode from Appstore) have
> a specific cost ?
>
>
> It's free. Although in Snow Leopard (1.6) you should download it from
> apple.com
>
> 4. How to check that whatever graphics card we have is compatible with
> CUDICA ?
>
>
> You should check your GPU device in the "System Information" application.
> Although, if you install CUDA 5.0 toolkit, you will get a CUDA item in the
> System Preferences pane.
>
> 5. As per a recent question on eeglab list, is it possible to use similar
> functionality to speed up other processes
> such as study precomputation in eeglab ?
>
>
> CUDAICA was developed to speed up the Infomax ICA algorithm developed by
> Sigurd Enghoff.
>
> I don't know about the study pre computation, but it all depends on the
> nature of the algorithm, as CUDA has some restrictive memory access
> patterns.
>
> Regards,
> Federico
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20121017/56356d5b/attachment.html>