On Nov 12, 2009, at 8:16 PM, Jon Elson wrote:

>> I'd love to see kinematics and/or TP calcs done on a GPU.  Recent  
>> video
>> cards do single precision floating point very quickly, and the newest
>> set of GPUs can do double precision natively as well.  Got a  
>> kinematics
>> problem that requires 1000 iterations to solve?  No problem,  
>> because 500
>> processors running at 700MHz can figure that out in a hurry.   
>> Concerned
>> about trajectory planning with massive lookahead while also taking  
>> into
>> account FO and other things?  No problem.  Run a bunch of scenarios  
>> in
>> parallel on the GPU and then pick the one that's closest to the  
>> actual
>> situation when the next servo cycle comes around.
>>
>> There are all sorts of things that can be done when you can (a) run a
>> lot of stuff in parallel and (b) throw away 99% of the work and  
>> still be
>> fast enough to control a machine.
>>
> The problem with GPUs is they are all proprietary (ie. different) from
> board to board,
> and manufacturer to mfr.  That would be a real problem.


That would be an advantage going with something like OpenCL.

To take advantage, one would have to chose a GPU that supported it.  
Just like gamers chose cards that support certain OpenGL extensions  
and such.

Heck, even if only a single GPU was supported, and I needed the  
capabilities it offered, chances are I would go with it.


Best,

Jeshua Lacock
Founder/Programmer
3DTOPO Incorporated
<http://3DTOPO.com>
Phone: 208.462.4171


------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Reply via email to