Just for my interest: MVC decoding usually contains a left and right eye stream, where one of the streams is a normal h264 stream. Which means that in theory MSDK could use hardware decoding for one eye and software decoding for the other eye. Is that the way it is implemented right now? Or are there technical problems with implementing such a mixed hw/sw decode?
madshi.net: that's a good point. And that should also apply to the encoder. There would be some technical problems though. An encoder can choose to predict the right view from the left (or is it left from right), although it rarely provides any compression benefits. If that is the case, to encode/decode the right view one would need fast access to the memory the left view is stored in. Cross GPU/system boundary memory accesses may be slow - just guessing though, don't really know.
The Media SDK codec is a bit more complex than the H.264 codec since it takes advantage of inter view (L,R) compression.
Let me clarify a bit more regarding HW acceleration of MVC codec. In connection with the release of our next generation platform, early next year, we will release a new driver. When this driver becomes available it can also be used on current generation platform (Sandy Bridge), effectively enabling MVC HW acceleration via Media SDK.
That said, dual H.264 decode (or encode) is always an option for S3D implementations. With current generation drivers this can be achieved using full HW acceleration (both left and right view).