Posted
Hello Andre,
Was thinking about picking up Synfire when I can afford it (got HN recently, and miss the mark for Synfire by about $100 at the moment..) but was talking with someone on another forum about it and the topic of processing came up...
I understand that it can take a while to process the data sets (no doubt there is a probably a pretty large set of neural network type processes and fitness functions cranking away under the hood) but I was wondering if GPGPU style processing would be worth investigating. Similar to the way protein folding programs were adapted to run on graphics processors to gain some 40x increase in processing speed (100-250x in the extreme cases).
I'm not sure how fit the data sets and algorithms are for this type of GPU processing, but it was just an idea. :)
Sat, 2009-01-31 - 16:39 Permalink
Hi Keith,
thanks for your thoughts. Yes, using the GPU for processing is a tempting idea. However, this woud require the algorithms and data structures used by Synfire to be converted and moved back and forth through an API bottleneck.
The current implementation is very lean and clean, deterministic and highly integrated with the musical knowledge base. I could rather think of forking off multiple processes to work on a separate instrument/track each to speed things up.
Andre