Closed Bug 1403091 Opened 7 years ago Closed 7 years ago

Rendering issues on Windows with Nvidia driver version 385.69

Categories

(Core :: Audio/Video: Playback, defect, P2)

defect

Tracking

()

RESOLVED WORKSFORME
Tracking Status
firefox55 --- wontfix
firefox56 --- wontfix
firefox57 --- wontfix
firefox58 --- affected

People

(Reporter: nical, Assigned: bwu)

References

Details

(Keywords: regression, Whiteboard: [gfx-noted])

This bug was reported by a few reddit users on this thread: https://www.reddit.com/r/firefox/comments/71xu5c/whats_the_state_of_hardware_acceleration_in_v55/ 

> Pages stop redrawing, meaning that occluded elements turn completely black
> when revealed again. After a second or so, the page turns completely black.
> Windows that aren't in focus don't redraw, so work until hovered over, at
> which point they turn completely black as well. Note that it's always black,
> not the background colour as shown in that post.
> 
> I'm attributing this to HA because it was off prior to 55, the update turned
> it on, and turning it off again has caused the issue to go away.


Rolling back to driver 385.41 fixed the issue.
Should we blocklist this driver version?  (385.69 on Fermi/GF110/GeForce GTX 1080)
Flags: needinfo?(jmuizelaar)
Priority: -- → P1
Whiteboard: [gfx-noted]
I've got another user here who seems to have the same issue.
https://www.reddit.com/r/firefox/comments/744hio/juddery_playback_on_youtube/

Can we just blocklist this and contact Nvidia?
On a sidenote, is there a way we can have a better system for detecting and preventing this type of thing in the first place?
A common complaint I see is that Firefox is not good on video. With some saying "57 is great, except for video."
Component: Graphics → Audio/Video
I'm assuming blocking for video decoding?
Flags: needinfo?(jmuizelaar)
Component: Audio/Video → Audio/Video: Playback
Keywords: regression
This doesn't sound like a video issue to me. Can we just blacklist layers on 385.41 for GTX 1080? It looks like it is just a buggy driver version.
Flags: needinfo?(milan)
The video issue was more from the comment 2 and the overall symptoms being that from a driver reset (or coming close to it.)  Together with comment 3, it was going into the "video causes a driver reset and then things look wrong everywhere".  That's a bit of a leap, agreed.
We could just block all acceleration on 385.69, but I'd feel better if we knew that a newer driver doesn't have the same problem.

From comment 0, 385.41 is good, 385.69 is bad and 387.92 is the latest - Caspy, nical, anybody knows if the latest one works?
Flags: needinfo?(nical.bugzilla)
Flags: needinfo?(milan)
Flags: needinfo?(caspy77)
Bas has a GTX 1080 GPU so you could ask him to investigate.
Flags: needinfo?(milan)
I've queried both redditors who reported this issue.
Waiting on responses.
Flags: needinfo?(caspy77)
Got one response so far: 

> It still happens but it's much less frequent to the point where I almost have to go out of my way to get it to happen by leaving a video running long enough or entering/existing fullscreen enough times.

Not exactly a glowing review of something that should Just Work®

===

As a sidenote I wonder how GPU manufactures are able to so frequently break their HWA for stuff.
The other responded saying:

> Haven't been getting any issues in several hours of usage.
We currently don't have the code to block Nvidia driver versions based on their versions (xxx.xx), but would need to expand the list into the yyyy.yyyy.yyyx.xxxx format where the last five numbers are the relevant ones.
So, this will be a larger patch, to add that capability (we already do something similar for Intel, and have the comparator enum for it), then actually do it.  Or, we'd need to figure out all the possible variations for the other numbers.  I'll try to get that from telemetry.

Anthony, do we have the ability to do the blocking of HW video decoding just based on the actual driver version (xxx.xx) for Nvidia?  I don't know if that would help here, but it may be worth it.
Flags: needinfo?(nical.bugzilla)
Flags: needinfo?(milan)
Flags: needinfo?(ajones)
Flags: needinfo?(milan)
Gerald,
Do you know the answer to comment 11?
Flags: needinfo?(gsquelart)
The D3D blacklists ("media.wmf.disable-d3d{9,11}-for-dlls") currently require the full DLL version, e.g.: "nvwgf2um.dll: 22.21.13.8253, 22.21.13.8233, ..."
So we'd also need more code if we wanted to support shorter driver version numbers.

I've always wanted to do that, but time flies, and these blacklists are now on their way out (see bug 1393392 and bug 1282335) so I think we shouldn't spend too much time on them.

However, if they are in fact the most convenient place to add such blocking for now, it would be simple enough to add, around http://searchfox.org/mozilla-central/rev/a984558fa2bbde6492d3fb918496fc0b0835b2ce/dom/media/platforms/wmf/WMFVideoMFTManager.cpp#379
Flags: needinfo?(gsquelart)
Just an update that the reddit user who reporting things being better in comment 9 on the most recent driver has rescinded that comment and says that the issue may in fact be worse.
I asked on reddit to see if the user is still experiencing the issue.
Flags: needinfo?(ajones)
The reddit thread seems to be dead. This issue isn't easily reproducible on 1060 GTX by my team with same drivers. Given that this is was reported against 55, it isn't worth holding up the 57 release for this.
I am going to check this bug.
Assignee: nobody → bwu
Close this bug since we have not seen any feedbacks or new reports from reddit for almost one month.
Status: NEW → RESOLVED
Closed: 7 years ago
Flags: needinfo?(milan)
Resolution: --- → WORKSFORME
I'm still experiencing the bug I reported, which I think is related: https://bugzilla.mozilla.org/show_bug.cgi?id=1409949

It's not consistent for me, it happens sometimes shortly after I turn my PC on. Sometimes it clears up after my PC has been running for a while, other times it just causes a BSOD. (And then may or may not re-occur when the PC is turned back on)

I saw something online saying that it could be caused by a NVIDIA's audio drivers, so after buying some separate speakers I've now disabled the NVIDIA audio devices, but the issue is still happening.
You need to log in before you can comment on or make changes to this bug.