User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:74.0) Gecko/20100101 Firefox/74.0

Steps to reproduce:

Install the Nvidia driver 445.75.
Surfing in the beb, working, etc. begins after 15-30min.
Betriebsystemname Microsoft Windows 10 Pro
Version 10.0.18363 Build 18363
Prozessor Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
Dell Inc. 1.18.0, 17.11.2019

Same Problem here:
https://www.reddit.com/r/firefox/comments/7xub85/weird_graphical_glitches_black_squareslines/

Actual results:

No readable content

Expected results:

Redable content

Could you attach the "about:support" output?

Could you try Firefox Nightly?

Also, when you hit this, it would help us to look at a GPU capture. Making it as simple as hitting Ctrl+Shift+3 (preferably in Nightly). The output is a "C:\Users\<your name>\AppData\Local\wr-capture" folder, which you'd need to compress and share with us (e.g. by sending me a link to Firefox Send)..

Setting to P2 for now, since it makes browsing barely usable on one of the popular HW configurations, at least until we know more.

Could you attach the "about:support" output? => yes
https://send.firefox.com/download/9ea528ecadcf7f3b/#h-K5415XfUNE4sV0EZin6g

Could you try Firefox Nightly? => yes, same problems here

Also, when you hit this, it would help us to look at a GPU capture. Making it as simple as hitting Ctrl+Shift+3 (preferably in Nightly). The output is a "C:\Users\<your name>\AppData\Local\wr-capture" folder, which you'd need to compress and share with us (e.g. by sending me a link to Firefox Send).. => Nothing happens, folder does not exist after i hit Ctrl+Shift+3 in Release and Nightly version.

Hey Rares,

Can you see if you can reproduce this bug? It looks like this user indeed has WebRender, they are not using DirectComposition. They have a NVIDIA GeForce GTX 1050.

Same here, a bit more up-to-date from this week, not like the link in my first post:
https://www.reddit.com/r/firefox/comments/fqafni/weird_graphic_glitches_in_firefox/

Today I worked with the nightly build all day, conclusion, the errors are still present here, but affect smaller areas or only sporadically in the picture, not like in the release version where you can hardly read the text.

(Firefox 74, Direct3D 11 (Advanced Layers), Same Nvidia driver version: 26.21.14.4575, 3-17-2020)

Can repro on ff webrender by forcing it to use my nvidia gpu.
Steps taken:

• upgraded driver
• Went to graphics settings and forced firefox to use nvidia gpu
• restarted firefox with nvidia gpu
• surfed for 10 minutes
• grabbed cookies and coffee (celebrations milk chocolate)
• unlocked machine
• Open firefox page is now garbled.

Chrome with nvidia GPU refuses to even start, just showing black screen after locking up machine for 1-2 seconds.

After restarting the machine, chrome started working again, but when switching tabs I often see flashes of garbled content.

https://www.nvidia.com/en-us/geforce/forums/game-ready-drivers/13/350427/nvidia-44575-is-broken/
-> Seems to also happen in other applications like Blender.

Rares, we don't need you to look at this anymore.

Andrew, we block the driver mentioned in https://bugzilla.mozilla.org/show_bug.cgi?id=1625857#c7?

I was able to reproduce this issue as well In the latest Nightly as well as the Latest Release versions, at first I couldnt reproduce the issue but You need to let it sit on a page for for quite some time, as long as I was scrolling reddit or changing tabs did issue did not occur at all, after leaving it idle for 10 minutes with Chrome in front view I was able to reproduce the issue, I also Noticed 2 Errors in the Failure log while scrolling reddit, I am not sure what causes them but after the second error I could notice this glitch right after.. Not sure if they are related.

Here are the following errors :

The only issue I noticed was the following error in the Failure log displayed multiple times on Nightly, im not sure what it means : GP+[GFX1-]: Present1 failed: 0x887a0001

In Release 74 there is this error displayed (GP+[GFX1-]: Updating unknown shared surface: 60129544279) as well as this one (CP+[GFX1-]: Failed to connect WebRenderBridgeChild) for scrolling really fast on reddit .

This issue started to occur as soon as I upgraded the driver version to the one from the reporter.

https://hg.mozilla.org/mozilla-central/rev/eb9466622b5c

Thanks for the report.
I filed a NV internal bug 2923118 to track this. I can report back if our engineers find something or are able to repro.

I could not immediately repro this with my own single-gpu GTX 1080 desktop and unfortunately at the moment I don’t have access to exactly similar systems the original reporters had.

From what I can see from the about:support attachments that I can access from the bugs, it appears that the reporters have dual GPU system with GTX 1050 or GTX 1060 and then an Intel GPU. So possibly they were using a notebook systems.

If you have time to spend on this in terms of engineering and QA, you may be able to narrow the blacklist based on the presence of NV+Intel setup (e.g. for dual gpu notebooks)..

Bugbug thinks this bug is a regression, but please revert this change in case of error.

Kimmo, it looks like Nvidia has a hotfix up: https://nvidia.custhelp.com/app/answers/detail/a_id/5009/track/AvMiwwriDv8S~eRBGvoe~yJl1Msq3y75Mv8S~zj~PP~b

So it seems like maybe this is not limited to dual gpu notebooks?

(In reply to Jessie [:jbonisteel] pls NI from comment #21)

Kimmo, it looks like Nvidia has a hotfix up: https://nvidia.custhelp.com/app/answers/detail/a_id/5009/track/AvMiwwriDv8S~eRBGvoe~yJl1Msq3y75Mv8S~zj~PP~b
So it seems like maybe this is not limited to dual gpu notebooks?

Ah, let me clarify: I meant that if there's more opportunity to do more QA on this, one avenue to test is whether this problem can be reproed on single-gpu systems or if this affects multi-gpu systems only. As it stands today, we on our end don't know the root cause of this particular bug, but we have not yet investigated this either. So my observation of reports of "dual-gpu" might be just a red herring.

The hotfix 445.78 talks about problems with image sharpening.
I could not repro the issue yet with 445.75 with image sharpening on single-gpu GTX 1080.

It would be useful if the original authors would have time try to repro with 445.78.

Hey Rares,

Can you confirm - the machine that you were able to repro this issue with, was it a dual-gpu machine?

Can you also download the hotfix here: https://nvidia.custhelp.com/app/answers/detail/a_id/5009/track/AvMiwwriDv8S~eRBGvoe~yJl1Msq3y75Mv8S~zj~PP~b and see if you could reproduce the original issue?

Hello Nils,

Can you try installing this hotfix from Nvidia: https://nvidia.custhelp.com/app/answers/detail/a_id/5009/track/AvMiwwriDv8S~eRBGvoe~yJl1Msq3y75Mv8S~zj~PP~b and let me know if you can reproduce any issues?

Hi, I've upgraded my Nvidia GeGorce GTX 1050 GPU driver to 445.78 and I was able to reproduce this issue in Firefox Release 74.0 as well as our latest Nightly 77.0a1 (2020-04-07).

Can you share the about:support for that machine you were using to test?

Added the about:support information from the G3 Laptop with Dual gpu's

Hi Kimmo, it does look like we can still reproduce the issue with 445.78 and a dual-GPU system. Haven't yet been able to find a single gpu system to try and repro there, but hopefully this can give some direction on your end?

Thanks for the reproductions. We have fixed an issue in drivers starting at 445.87 which probably is the exact same issue discussed here.

It would be great if the original authors could confirm if 445.87 fixes the problem? The drivers can be downloaded from https://www.nvidia.com/download/find.aspx

Hello Nils - if you get a chance, can you try downloading the driver mentioned in https://bugzilla.mozilla.org/show_bug.cgi?id=1625857#c31 and see if that fixes your issue?

Just NI-ing rares to try out new driver

Kimmo, can you give us so more information about what the problem was? Is there a way that we can work around it? Do you suggest we block the old one?

FYI, I got a response back from bug 1627461 who said that rolling back the Nvidia driver to 442.59 resolved the issue for them.

Kimmo, can you give us so more information about what the problem was? Is there a way that we can work around it?

The corruption occurs when computers come back from S3/S4 sleep. The issue affects "Pascal" generation of chips (https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_10_series).

At the time there's no known working workaround.

Do you suggest we block the old one?

I think it makes sense to block the driver at least on laptops, where users trigger the sleep frequently.

The problem itself also manifests with desktops, if the users use the sleep functionality -- thus it might be warranted to block there too. If you have more user data about whether or not the sleep functionality is generally used by desktop users, you might use that info to guide the decision.

Taking into account that the exact same bug has been reported in Opera, I believe it’s more likely to be a bug with Nvidia rather than Firefox.

I can't seem to reproduce this issue anymore with their latest update version / downgrade version 442.59.

Rares, were you able to try out the driver mentioned here: https://bugzilla.mozilla.org/show_bug.cgi?id=1625857#c31 version 445.87?

Nope I couldnt reproduce the issue with version 445.87 either. I tried Release 75, Beta 76.0b6 as well as todays Nightly.

NVIDIA GPU drivers version 445.87 are finally bug free.
Version 445.75 and version 445.78 (mentioned in bug #1627428) have to be added to block list.

These are current, open bugs with a Severity of critical. The Severity of these bugs is being changed to S2 to be consistent with the May 4 2020 Severity definitions.

Please let Release Management know if these bugs are still S2.

I think we can make this an s3. Nvidia released a driver that fixes this and we don't want to block users from hardware acceleration.

We're just going to leave this as is and hope people update there driver if they have problems.