(no subject)

[torkell]

It's rather crazy that despite Flash being completely the wrong tool for playing videos, it does a better job than whatever Firefox uses for HTML5 video.

It's very crazy that on the modern Internet the state-of-the-art for video playback is so much worse than what any half-decent media player could do 15 years ago. Seriously, are DXVA and video overlays still a thing, or does everything just throw all the pixels at the CPU and hope that it's fast enough?

Comments

[crschmidt.livejournal.com]

Mostly the latter. I know that Chrome will do the former on many graphics cards, but graphics cards so So. Fucking. Shitty. and give you no visibility into whether the bits you send in are giving you video, or a mash of green pixels, or just a blank black screen; or whether they're going to drop 83% of the frames because you dared to try to play 1080p with a non-baseline H.264 profile; and at the web level, you just have no knowledge with which to make decisions on this level. (There is actually a browser API for getting graphics card models, but it requires creating a webgl context; creating a webgl context on its own is enough to really screw the pooch for many users.)

I expect that nightly Firefox is probably noticably better. They're doing uplifts from 43 to 42 for a significant chunk of the MSE improvements. We put off switching YouTube to using the HTML5 player in Firefox for more than a year hoping Firefox would get their act together, and eventually it was fixed-enough that it was time to go ahead and make the switch even though everything is still terrible.

Chrome is probably noticeably better than Firefox.

[torkell]

I thought as much... it's one of those things where the increase in CPU power means that going "sod it" and doing everything on the CPU is a practical solution. I know I've written code where the fact that it's using some incredibly stupid algorithm doesn't actually matter because the underlying hardware is fast enough. It just irks me because I know a Pentium M 1.5GHz, despite being woefully underpowered these days, is actually fast enough for HD video playback if the software knows how to do so.

On graphics cards: I'm not surprised - graphics drivers seem to be one of the less stable bits of software I've dealt with over the years (along with video codecs which seem to find equally hilarious ways to fail). One part of $WORK involves trying to make Android's video codec stack behave and finding such fun bugs as it only rendering the first frame of a video stream on some phones. Except when it works fine, but later deadlocks when releasing the codec if at any point you decided not to draw some frames.

I may have to give Firefox Nightly or Alpha Developer Edition a try, though I'm a bit reluctant to do as a colleague's install of Nightly regularly detonates all his tabs (despite e10s being enabled - or possibly because it's enabled). Though actually what I've decided to do with the laptop is stick Windows 7 on it, since that brings DXVA2 and WMF support which I think is what Firefox is actually targeting for video playback.

[torkell]

...and give you no visibility into whether the bits you send in are giving you video, or a mash of green pixels, or just a blank black screen;

You forgot "or crash the graphics driver and take out the browser for good measure because Fuck You that's why". Sadly that's on the shiny new GeForce 950 GTX in the desktop, not the laptop which at least has the excuse of being ancient.

I swear, it all reminds me of the system I had back at university, and the amount of hassle I had trying to make that doing supposedly simple things like "play a MP3 without stuttering" (which should be trivial with an Athlon 1.3, but becomes near-impossible when using the VIA 686B chipset and a SB Live - the two did not get on at all).

[tau-iota-mu-c.livejournal.com]

Thank goodness for the existence of youtube-dl.