Warehouse-scale video acceleration: co-design and deployment in the wild | Hacker News Reader

Warehouse-scale video acceleration: co-design and deployment in the wild

dl.acm.org

33 pointsjsnell9 comments

The Youtube blog post is here: https://blog.youtube/inside-youtube/new-era-video-infrastruc...

i recently read the the firesim from the berkeley arch group and that was interesting and impressive so I'll eventually read this too. but can talk about the fact that there are 51 authors on this paper? i was annoyed when i had to include 10 other authors on my last paper (most of whom did nothing material) for political reasons. what is the point? if everyone only really counts first author papers then who cares? this kind of stuff leads to citation/publication inflation

That was the thing I notice as well. I guess it is some sort of Resume -Driven Development?

I think it's fantastic that all contribution are included.

So much better than the politics of exclusion that often mean that mere (/s) programmers who implemented something never get acknowledged.

not sure what publications you've been reading but every one i've written and every one anyone in my lab has written includes the research engineers (ie devs) in the author list

>The high compute cost of VP9 makes it computationally infeasible at scale in software.

I felt this pretty much sums up my thought on the paper. It was specifically designed for VP9. Given how the VPU Encoder Core ended up having the same speed in encoding both H.264 and VP9. Despite H.264 using around one fifth of time on Software CPU encode.

And also the reason I think why AV1 is widely deployed on Youtube. Although Google and Youtube repeatedly refuse this claim.

@jsnell Please Tell me you saw this paper on Doom9 :)

> Please Tell me you saw this paper on Doom9

Afraid not, I'm not really a video codec enthusiast, my interest in the paper was from the systems design perspective :) Is there some discussion on the paper there?

>Is there some discussion on the paper there?

Not yet, And likely not since they are video codec enthusiast. I just happen to submit the same thing on Doom9 a moment before this appeared on HN so I was wondering.

See also FBs very different approach: https://engineering.fb.com/2021/04/05/video-engineering/how-...

Reply on news.ycombinator.com