2 ms·
A lost art to be rediscovered. This type of insane hyper-optimization has a lot of applications today. Even in the exact domain of video codecs on modern system
by bob1029 7y ago
A lost art to be rediscovered. This type of insane hyper-optimization has a lot of applications today. Even in the exact domain of video codecs on modern systems.
Perhaps we cannot get much better perceived quality per bit, but I would strongly argue that we could improve other areas such as encoder/decoder latency.
I've got a little toy project where I am generating 1080p 24bpp frames in a C#/AspNetCore webservice and sending them to client browsers at 60fps over websocket. Nothing serious, I was just trying to test the limits of what I could do with current-year managed language and some crappy fun code. Think of it as a really bad attempt at Google Stadia. I currently see about 5 frames of latency with MPEG-1 using bidirectional pipes against FFMPEG. While this is an extremely encouraging first result, I feel there is some compromise I can make on quality+bandwidth in order to further reduce this frame latency (I.e. if I were to hand-roll some codec for this specific application).
If I could get the frame latency down to the ~30ms range on localhost, this opens the door really wide on some crazy project ideas I've had regarding novel real-time graphics approaches. I will definitely be printing this article off for deeper review.