1085673 - Visual glitch semi-randomly renders vertical bars in browser chrome with OMTC+D3D11 enabled (NVIDIA GeForce GTX 560 Ti)

Status: UNCONFIRMED

(Core :: Graphics, defect)

Description

[oo.rio.oo]

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:33.0) Gecko/20100101 Firefox/33.0
Build ID: 20141011015303

Steps to reproduce:

Launch Firefox after installing the 33 release channel update


Actual results:

Firefox randomly renders odd vertical 'bars' or 'stripes' at semi-random positions in the browser chrome.

These positions seem to correspond with cut points on the edge of or at logical points inside of certain UI elements: edges of buttons on the bookmarks bar; trailing before or after the text in the address bar; at the cutoff point where a tab transitions from its straight section into the slanted left and right end points; etc.

At times it also shows up randomly in the top 'empty' space of the tab bar or randomly across the home button. (Though that may just be a lot of cut points for a complex UI widget, really can't tell...)

At other times (and this is particularly weird) one of the lines syncs up with the text caret in the address bar and starts blinking in the same rhythm.

The issues cease when hardware accelerated rendering is disabled, but Firefox has always behaved fine before with it enabled. Seems like 33 started tripping something that the nVidia driver doesn't like.


Expected results:

Firefox's UI should not exhibit these weird graphical anomalies.

Comment 1

[oo.rio.oo]

See also: https://support.mozilla.org/en-US/questions/1015729
Where a user hooligan131 may be describing something related.

And: https://support.mozilla.org/en-US/questions/1025786
Where user M.G. is describing the exact same problem.

Comment 2

[Loic]

Could you attach some screenshots, please.

Comment 3

[oo.rio.oo]

Attachment: Shows glitch occuring in tab area

Comment 4

[oo.rio.oo]

Attachment: Shows glitch occuring in bookmarks bar

Comment 5

[oo.rio.oo]

Attachment: Shows glitch occuring in both bookmarks bar and start of address bar

Comment 6

[Loic]

Could you type about:support and post the "graphics" section, please.

Does it disappear if you restart Firefox?

Comment 7

[oo.rio.oo]

As requested; I attached some screenshots marking glitches as they occured.

I've seen other, more obvious occurrences, but it's quite hard to catch them in a screenshot as they tend to disappear if a repaint is triggered. Sometimes even when a repaint in a seemingly completely unrelated area of the screen happens. E.g. the glitch seen in the tab area disappeared the moment my mouse left the area of the bookmark button you can see in its hover-state on the screenshot in question (attachment 8508846 [details]).


Also, as requested; the information from my graphics tab:


Graphics
========
Adapter Description        | NVIDIA GeForce GTX 560 Ti
Adapter Drivers            | nvd3dumx,nvwgf2umx,nvwgf2umx nvd3dum,nvwgf2um,nvwgf2um
Adapter RAM                | 1023
Device ID                  | 0x1200
Direct2D Enabled           | true
DirectWrite Enabled        | true (6.1.7601.18245)
Driver Date                | 3-14-2013
Driver Version             | 9.18.13.1422
GPU #2 Active              | false
GPU Accelerated Windows    | 1/1 Direct3D 11 (OMTC)
Vendor ID                  | 0x10de
WebGL Renderer             | Google Inc. -- ANGLE (NVIDIA GeForce GTX 560 Ti Direct3D9Ex vs_3_0 ps_3_0)
windowLayerManagerRemote   | true
AzureCanvasBackend         | direct2d
AzureContentBackend        | direct2d
AzureFallbackCanvasBackend | cairo
AzureSkiaAccelerated       | 0


While the driver from 3-14-2013 is a bit older, it is the last known good driver, i.e.; free of random BSODs and driver resets, for Geforce 400/500 series cards that nVidia released.

Comment 8

[oo.rio.oo]

Also; the issue does not disappear if Firefox is restarted; the issue will appear and disappear semi-randomly while the browser is being used.

Disabling hardware acceleration prevents it from occuring. Something in hardware-accelerated rendering that was changed between FF32 and FF33 is causing this, it seems.

Comment 9

[Loic]

So it's related to graphics.

The 1st step is to update your drivers to the last ones:
http://www.nvidia.com/download/driverResults.aspx/77837/en-us (check the language)

After, that if the issue is still here, try with OMTC disabled:
about:config > layers.offmainthreadcomposition.enabled = false (restart FF to apply)

Comment 10

[oo.rio.oo]

Attachment: Left edge pixel restored

Comment 11

[oo.rio.oo]

I can't upgrade beyond this driver revision (314.22) due to stability problems with later drivers. While the current driver release (340.52) is better, it still does not solve the random driver freeze+recovery or BSODs that all later drivers tend to exhibit on a wide range of Geforce 400/500 series hardware.

Disabling the layers.offmainthreadcomposition.enabled setting does work; it prevents these random glitches completely. And it also seems to fix the 1px wide trim on the left hand side. Compare attachment 8508902 [details] with the others for reference.

Comment 12

[Loic]

Nical, do you know a possible workaround to not disable OMTC completely?

Comment 13

[Nicolas Silva [:nical]]

Could you try to fallback to D3D9 compositing? To do this, switch OMTC back on, and set the pref "layers.prefer-d3d9" to true in about:config.

Comment 14

[Milan Sreckovic [:milan] (needinfo for best results)]

(In reply to Nicolas Silva [:nical] from comment #13)
> Could you try to fallback to D3D9 compositing? To do this, switch OMTC back
> on, and set the pref "layers.prefer-d3d9" to true in about:config.

And restart.

Comment 15

[NVD]

http://www.geforce.com/whats-new/articles/geforce-344-48-whql-driver-released

Please update to GeForce 344.48 WHQL driver and don't say you can't update.

Comment 16

[oo.rio.oo]

(In reply to Nicolas Silva [:nical] from comment #13)
> Could you try to fallback to D3D9 compositing? To do this, switch OMTC back
> on, and set the pref "layers.prefer-d3d9" to true in about:config.

Falling back to D3D9 also prevents the visual glitches, but it also disables DirectWrite, making text fidelity poor in comparison. (The difference is quite jarring moving away from DirectWrite; almost hurts the eyes.)
I think you'd rather have the reduced performance of going without OMTC than the reduced text fidelity of giving up DirectWrite along with the reduced performance of giving up Direct2D.

This little test atleast tells you that the problem is in the D3D11 OMTC bits, I guess?


(In reply to NVD from comment #15)
> http://www.geforce.com/whats-new/articles/geforce-344-48-whql-driver-released
> 
> Please update to GeForce 344.48 WHQL driver and don't say you can't update.

WHQL or not, drivers above 314.22 all exhibit varying degrees of stability problems with Geforce 400/500 series leading to graphics freezes and driver resets or unrecoverable freezes, BSODs and data loss. In some extreme cases nVidia's fiddling with voltage and clock tuning allegedly even caused irrepairable hardware damage.

I'd rather not roll the dice on the card in this system being one of those (un)lucky ones.

Comment 17

[Nicolas Silva [:nical]]

(In reply to oo.rio.oo from comment #16)
> This little test atleast tells you that the problem is in the D3D11 OMTC
> bits, I guess?

Yes, that's what I was looking for.

Comment 18

[65feb6b8c9726133b18ac2f2ac26e8bc]

> WHQL or not, drivers above 314.22 all exhibit varying degrees of stability problems
> with Geforce 400/500 series leading to graphics freezes and driver resets or
> unrecoverable freezes, BSODs and data loss
Yeah, that's true. I also use 314 because of random BSODs with all newest versions I tried

Comment 19

[Gregory Pappas [:gregp]]

Reporter, are you still experiencing this issue?