1728169 - High VRAM clocks when playing videos (Twitch/Youtube) only on FF

Status: UNCONFIRMED

(Core :: Graphics: WebRender, defect, P3)

Description

[FFpr]

Attachment: 122.png

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:92.0) Gecko/20100101 Firefox/92.0

Steps to reproduce:

Play a video on Twitch or Youtube.

Actual results:

As you can see from the attached screenshot, on the same 1080p/60fps Twitch livestream, when using Firefox the VRAM clocks oscillate a lot and reach almost 1000MHz (lower when safe mode is on) causing the GPU to heat up (top of the graph), eventually and routinely turning on my GPU fans when exceeding 60C.

On Edge or Chrome the VRAM remains idle at 200MHz with the occasional spike allowing the GPU to sit at around 50C indefinitely.

Ryzen 3700X - RX 5700 XT (Adrenalin 21.6.1 and 21.8.2)

Expected results:

VRAM clocks shouldn't be that active causing the GPU to warm up that much.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The Bugbug bot thinks this bug should belong to the 'Core::Audio/Video: Playback' component, and is moving the bug to that component. Please revert this change in case you think the bot is wrong.

Comment 2

[John Lin [:jhlin][:jolin]]

Thanks for reporting this issue.

I've tried it on my Windows 10 machine (GPU: NVIDIA Quadro P400) and found the GPU memory clock goes up to ~1000Mhz for several seconds when videos (e.g., [1]) start to play then downs to ~200Mhz afterward until I switched to another video. Could you please try playing it and see if your VRAM clock keeps high the whole time? If not, please share a Youtube video link that you observed the high clock frequency issue.

Also, could you please upload the troubleshooting info[2] too? Thanks a lot!

[1] https://www.youtube.com/watch?v=LXb3EKWsInQ
[2] https://support.mozilla.org/en-US/kb/use-troubleshooting-information-page-fix-firefox#w_accessing-the-troubleshooting-information-page

Comment 3

[FFpr]

Attachment: troubleshooting.txt

Comment 4

[FFpr]

Attachment: YouTube.png

That video displays the same issue, I've let it run to completion at 1080p/60fps for consistency.

Comment 5

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

We rely on the OS to provide hardware decoders, so if those are the cause, then we have limited ability to impact this. However, if this is elsewhere in the stack we may be able to do more. We'd expect the GPU usage to go up when it's decoding video, but if we're using a lot more than something a Chromium based browser or even Firefox in safe mode, it would be useful to figure out why.

Could you see if the same issue happens when using a new profile in Firefox? This would help check if certain things being disabled by safe mode are contributing to the lower usage there.

Could you also link the specific videos you're testing with? It would be helpful to make sure we're reproducing in the same way.

Comment 6

[FFpr]

Attachment: GPUZcomp.png

I'm using the same video linked above forcing 1080p/60fps:
https://www.youtube.com/watch?v=LXb3EKWsInQ

Here's a comparison of roughly the first minute of the video using GPU-Z logs between Firefox on a new profile and Chrome.

Firefox is red, Chrome is green (Chrome didn't have ads so the log is a few seconds shorter)

Comment 7

[Bryce Seager van Dyk [:bryce] (he/him) - Not reading bugmail]

Thanks.

For posterity, the tools used above are

• GPU-Z
• HWInfo log viewer

For the video in comment 2 I see YT serve me VP9 for both videos. In both cases I believe Blink and Gecko will use a Windows decoder (something like D3D11VideoDecoder). It's possible there's some decoder config which is impacting Fx here. It could also be our rendering pipeline is doing something differently.

Comment 8

[FFpr]

Attachment: GPUZcompAfter21.9.1.png

Today AMD released new Radeon drivers, 21.9.1, and I noticed they addressed some 5500XT memory clock issues while idle, so I wanted to check if they changed anything on my 5700XT since they're both RDNA 1 and yes, it might be a coincidence but it seems fixed now, VRAM now stays at 200MHz as expected after an initial spike.

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-21-9-1

Attached is a new run of 2 minutes of the same video in comment 2 at 1080p/60fps.
FF (new profile) is red, Chrome is green.

Memory clock isn't included because it was just a flat line at 200MHz on both browsers, but you can see that GPU clock, load and power draw are still higher than Chrome's.

Comment 9

[Paul Adenot (:padenot)]

Jeff, probably something of interest for you, for video playback power usage on Windows with an AMD GPU. Not sure where the problems lies yet though.