It’s not stateless end-to-end, it just means the client needs to keep track and pass the state rather than drivers or hardware
I’m not 100% on the motivation but from an architectural standpoint it does make sense - your software can now do many new and weird things without a hardware change
One example I saw was allowing an arbitrary number of streams to be processed simultaneously, just passing the different context state for each stream
sounds nice but what is the advantage of a stateless decoder?
How can it be stateless if you need information from past I frames?
It’s not stateless end-to-end, it just means the client needs to keep track and pass the state rather than drivers or hardware
I’m not 100% on the motivation but from an architectural standpoint it does make sense - your software can now do many new and weird things without a hardware change
One example I saw was allowing an arbitrary number of streams to be processed simultaneously, just passing the different context state for each stream