[Developers] newfmin on GPU with aribitrary number of variables

[dave fournier]

On 12-05-09 10:45 AM, Ian Taylor wrote:

The BFGS without GPU is already in the version of newfmin.cpp.  Look for 
something like
gpu _flag   near label21:



> Thanks Dave.
>
> Just in case anyone else is as ignorant as I was yesterday, don't 
> waste your time trying setting up the GPU stuff on a virtual machine, 
> because it won't work.
>
> That got me thinking it would be nice to work toward having 
> independent options for the BFGS algorithm and GPU calcs. Dave did 
> note that the characteristics of the BFGS algorithm make it well 
> suited to GPU, but it's easy to imagine that there could be cases 
> where the new algorithm would perform better even without the extra power.
> -Ian
>
> On Wed, May 9, 2012 at 7:50 AM, dave fournier <davef at otter-rsch.com 
> <mailto:davef at otter-rsch.com>> wrote:
>
>
>     To get the flexibility for testing the GPU version of newfmin.cpp
>     I have modified it to work with vectors not divisible by 256.
>     GPU's like powers of 2. For example they can do 256 or perhaps 4*256
>     operations concurrently.  New version is attached.
>
>
>
>     b-project.org/mailman/listinfo/developers
>     <http://b-project.org/mailman/listinfo/developers>
>
>         _______________________________________________
>         Developers mailing list
>         Developers at admb-project.org <mailto:Developers at admb-project.org>
>         http://lists.admb-project.org/mailman/listinfo/developers
>
>
>
>     _______________________________________________
>     Developers mailing list
>     Developers at admb-project.org <mailto:Developers at admb-project.org>
>     http://lists.admb-project.org/mailman/listinfo/developers
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/developers/attachments/20120509/8d2ac192/attachment-0001.html>