Use a graphics card for Geoprocessing
Seems like more geographers should be following the efforts of people finding ways to use the graphics card for general programming. I had always thought you could only do integer operations on the graphics processor, but here‘s an interesting paper describing “float-float” operations. I suppose this means one could handle double precision geometries from a 9.2 geodatabase. I wonder how much faster building topology could be done on the gp. Maybe there could be an option on the topology toolbar that would allow me to leverage the graphics card. Granted each graphics card is different, but it should be possible to write adapters to abstract away the differences.
The Memory Leak 
this is an intersting link describing a general programming wrapper around the GPU pixel shader pipeline. It’s from MS Research, uses .Net and to be honest, looks like it could be very helpful in some geoprocessing routines…
channel 9 video
http://channel9.msdn.com/ShowPost.aspx?PostID=229585
whitepaper:
ftp://ftp.research.microsoft.com/pub/tr/TR-2005-184.pdf
cheers
brian
Good to hear from you Brian! Does this mean you’re going to program your new xBox to do some real work? 🙂
Thanks, for the links, this looks promising.
To me the fuzzy part is parallelizing the algorithm. I wish I had time to try adapting this point in polygon algorithm http://aprsworld.net/info/mpi/paper.pdf#search=%22parallelization%20gis%22 so that instead of using multiple cpus, it just uses a single gpu, via the Accelerator.
Regards, Kirk
[…] I really like the in-memory FeatureGraphicsLayer in 9.2 – very handy indeed. In 9.3 I’d like to see something similar for raster layers. Also, provide classes similar to all the RasterXXOp classes in ESRI.ArcGIS.GeoAnalyst and make them available the web adf. Instead of operating on file-based rasters though, they would need to operate on RasterGraphicsLayers. Also, maybe the SOC could be made smart enough to use the GPU for the raster processing of RasterGraphicsLayers. […]