View Single Post
Old 25th June 2017, 02:43   #235  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Ah, when talking about parallelism, I meant the encoder rather than the decoder, which is usually more compute intensive and seems to increase in complexity sometimes an order of magnitude or more per generation, while the decoder seems to be only a modest increase albeit still enough to hurt in the current climate of slow single thread improvement.

It seems that the likes of Netflix and Google get around parallelisation limits (atleast limits that dont impact compression efficiency) of H264, HEVC and VP9 by employing chunked encoding over the gamut of a server farm/datacenter, but this is still limited to CPU compute.

It would be interesting to see if we could at least see GPGPU employed efficiently for chunked encoding, if not for full length videos.
soresu is offline   Reply With Quote