YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]
-
Hello all!
I'm wondering if anyone has been getting artefact like screen flickering on Vivaldi 6.1 while watching videos on YouTube?
I've narrowed this down to only happening in Vivaldi 6.1 and I think it's because of the process name spoofing causing AMD drivers to load chrome optimisations which are incompatible. I came to this conclusion because of this thread.
Testing the Vivaldi snapshot prior to 6.1 as a standalone install doesn't display the problem.
Question: Does anyone know if there's a flag for Vivaldi 6.1 to disable the process name spoofing so that I can test this on and off within the same Vivaldi version?
I realise I can disable hardware acceleration, but I don't want to, I want to conclusively test that the driver optimisations applied meant for chrome in 6.1 are causing these issues and that my GPU isn't at fault.
Thanks for reading!
Problem Description & Background Information
- The flickering looks like green / red / white garbled screen flashing artefacts as you would get when a GPU is dying.
- It seems to only occur in AVC1 encoded youtube videos such as this.
- Flickering occurs occasionally and randomly but most prominent between 1min and 1:40 mark on that video, but I find sometimes you need to skip to the end and then back to that time stamp to kick it off.
- Flickering affects whole screen including windows task bar, it's not just corrupting the video.
- Running youtube in "theatre mode" if that makes a difference.
Troubleshooting Performed
- Issue doesn't occur in Firefox portable or Chrome portable.
- Issue doesn't occur in standalone install of Vivaldi.6.0.2979.22.x64.
- Issue doesn't occur on desktop, at idle or in gaming or any other video playback (for instance MediaPlayerClassic-HC).
- Issue doesn't occur on other youtube videos without AVC1 encoding.
- Tried to rollback AMD drivers using DDU (driver removal utility), and issue occurs on 23.5.3 and 23.4.3.
- Issue doesn't occur when I try to record a video of the problem using AMD screen recorder, I think this is because the GPU goes into high power state when doing that and it alleviates the issue.
- Ruled out my screen. The artefacting doesn't cover the OSD if I bring it up.
- Ruled out my DP cable. The artefacting doesn't happen during gaming or windows desktop at idle, etc.
System
Vivaldi: 6.1.3035.75 (Stable channel) (64-bit)
Windows 10 64bit Pro: 10.0.19045 Build 19045
AMD GPU: Radeon RX580
AMD Drivers: 23.5.2 -
@Pulseammo Hello and Welcome to the Vivaldi Community
I can't test this directly myself, as I have Nvidia. I can only give some general advice. I think it's safe to assume that Vivaldi has many users with AMD and many of them also use YouTube regularly. So I would expect more forum reports if this was a general issue, but you never know.
Question: Does anyone know if there's a flag for Vivaldi 6.1 to disable the process name spoofing so that I can test this on and off within the same Vivaldi version?
No, this is built into the executable itself.
I realise I can disable hardware acceleration, but I don't want to
This is only for testing. If the issue disappears with HW accel. off, it's a good indicator that it's the GPU drivers causing it.
I'd also recommend you have a good look at your AMD control panel, and reset any global settings or any set specifically for Vivaldi (if they exist).
Issue doesn't occur in standalone install of Vivaldi.6.0.2979.22.x64.
That's a good test, and might indicate a problem in 6.1 with AMD drivers. It might also mean your standalone install is a completely clean profile. Make sure to try a clean standalone install of 6.1 as well.
Also make sure of course to have followed the full troubleshooting guide to the end:
https://help.vivaldi.com/desktop/troubleshoot/troubleshooting-issues/Some more tips:
First make sure you have no experimental flags set in
vivaldi://flags
Any flag changed from default will be at the top of the list and outlined. Click reset all and restart the browser.After disabling HW accel and the issue goes away, enable it again and try to toggle some GPU-related flags. Make sure to toggle one at a time, restart the browser and test again. Some that might be relevant:
chrome://flags/#disable-accelerated-video-decode
chrome://flags/#disable-accelerated-2d-canvas
chrome://flags/#enable-gpu-rasterization
chrome://flags/#use-angle(test different backends)Try to download the video with yt-dlp, then open it directly in Vivaldi. It should be sufficient to just download the 1080p AVC1 video without audio (i.e. no need for ffmpeg)
yt-dlp.exe -f 137 https://www.youtube.com/watch?v=GejprXndyLk -
@Pathduck said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
@Pulseammo Hello and Welcome to the Vivaldi Community
Thanks for the welcome, and a bigger thanks for your reply, some of your suggestions have brought up some new information. I've also recorded a video of the issue, I managed to catch it at one of the more extreme examples.
I can't test this directly myself, as I have Nvidia. I can only give some general advice. I think it's safe to assume that Vivaldi has many users with AMD and many of them also use YouTube regularly. So I would expect more forum reports if this was a general issue, but you never know.
Yes, I can't rule out it's not just my GPU dying, but perhaps others will post now that I've mentioned it.
This is only for testing. If the issue disappears with HW accel. off, it's a good indicator that it's the GPU drivers causing it.
Tested this: Problem doesn't occur, so it seems to be GPU or Drivers related.
I'd also recommend you have a good look at your AMD control panel, and reset any global settings or any set specifically for Vivaldi (if they exist).
Tested this: Drivers are a clean install at default settings, I can't see anything relating to Vivaldi in there.
Issue doesn't occur in standalone install of Vivaldi.6.0.2979.22.x64.
That's a good test, and might indicate a problem in 6.1 with AMD drivers. It might also mean your standalone install is a completely clean profile. Make sure to try a clean standalone install of 6.1 as well.
Ok here's where it gets interesting, I stand corrected here.
6.1 Standalone on a clean profile is just as bad as my permanent install, I can set off the flickering somewhat easily by starting the video and clicking the seek bar at 1min30 mark, which results in multiple coloured flickers.
I decided to retest Vivaldi.6.0.2979.22.x64 standalone and I found the flicker does actually occur, it's extremely infrequent, maybe one singular flicker in 10 minutes of testing, I almost thought I had imagined it.
This then lead me to re-test Chrome, this time instead of using a portable app I installed Version 114.0.5735.134 (Official Build) (64-bit) and I found the bug to occur at the same rate as Vivaldi 6.0. (A single flicker every 10 minutes of testing).
I re-tested Firefox, this time as an installed full version and the problem doesn't occur at all, same as the portable version.
First make sure you have no experimental flags set in
None set.
chrome://flags/#disable-accelerated-video-decodeTested: Seems to fix the issue.
chrome://flags/#disable-accelerated-2d-canvasTested: Doesn't fix the issue.
chrome://flags/#enable-gpu-rasterizationTested: Seems to fix the issue.
chrome://flags/#use-angle(test different backends)Tested: OGL & DX9 seems to fix the issue, DX11 issue is visible. DX11on12 issue is visible but maybe reduced a bit in frequency.
Try to download the video with yt-dlp
Tested: Tried for about 10minutes clicking about the seek bar to trigger the issue and couldn't get it to occur when the video was downloaded using 6.1. There does seem to be something about clicking about in the seek bar that sets it off in youtube as opposed to just watching the video start to end.
Conclusion?
It would seem that since this affects Chrome then it's an upstream bug, but Vivaldi 6.1 seems to display it much, much more readily.I've recorded a video of the issue using my phone. Basically when I click the 1min30 mark on the seek bar it seems likely to trigger the issue, once the issue has started then clicking back to the same area will keep the artefacts occurring almost like whatever corruption occurred is trapped in a buffer. If I click further on in the video and then click back to 1m30 it sometimes plays cleanly, sometimes not.
Normally I'd just say this type of flickering is 100% a hardware fault but I've still been unable to re-create the issue anywhere else outside of 2 YouTube videos that I've noticed it on. It's possible my GPU is just failing and the trigger is low power state combined with video playback since it seems to be stable in high power state & the same video playback (for instance if I load a GPU heavy application in the background and then play the video). There was a known AMD driver bug for flickering during video playback, but I think it was in 2022, I wonder if it's a hangover from that issue.
I'll take a look in the chromium issue tracker and the chrome one if it exists. In the mean time if anyone else has this issue please post about it!
-
@Pulseammo Oh yeah, I can get why that is disturbing!
Thanks for investigating, the detailed reply and especially for testing in Chrome, as it shows it might be a Chromium (upstream) issue.
chrome://flags/#disable-accelerated-video-decode
Tested: Seems to fix the issue.
chrome://flags/#enable-gpu-rasterization
Tested: Seems to fix the issue.Good test - what about only setting enable-gpu-rasterization to Disabled, while leaving accelerated-video-decode on (Default)? Not actually sure what's the Default here, but assuming it's Enabled...
chrome://flags/#use-angle (test different backends)
Tested: OGL & DX9 seems to fix the issue, DX11 issue is visible. DX11on12 issue is visible but maybe reduced a bit in frequency.Also interesting. DX11 then. More details is always good. Too bad DX11 is kind of the go-to renderer for performance reasons so changing is not recommended.
Try to download the video with yt-dlp
Tested: Tried for about 10minutes clicking about the seek bar to trigger the issue and couldn't get it to occur when the video was downloaded using 6.1. There does seem to be something about clicking about in the seek bar that sets it off in youtube as opposed to just watching the video start to end.OK might not be the same when directly playing back an MP4 file.
Does it occur on other sites or only YouTube? Vimeo?
Try googling for "AVC1 test video" - here's a classic test:
https://test-videos.co.uk/bigbuckbunny/mp4-av1Here's a good internal status page showing GPU details as well:
vivaldi://gpu/Unfortunately the upstream bug tracker for Chromium is very tech-heavy and requires a Google account but try to search if you can:
https://bugs.chromium.org/p/chromium/issues/listIf you want to go really down the rabbit hole, various Chromium builds are available here:
https://chromium.woolyss.com
From current to bleeding edge, and older versions are on the Github. Usually "Archive" is the best. Make sure to delete User Data when testing different builds.Sometimes it might be an idea to test with newer releases, possibly the issue is fixed say in v. 116.
-
I also have an AMD Radeon RX 580, however my current driver version is the latest one from Windows update (31.0.12027.9001, which is from late March) and I haven't experienced any artifacts or other graphical issues after the update to 6.1.
-
@Pathduck said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
Good test - what about only setting enable-gpu-rasterization to Disabled, while leaving accelerated-video-decode on (Default)? Not actually sure what's the Default here, but assuming it's Enabled...
I tested each of the flags in isolation. When rasterization was disabled hardware decoding still occurred according to the task manager graphs (there's one for GPU usage and a separate one for video decode). It's possible one of my results is a false negative, the issue doesn't reproduce in a clean and consistent manner.
I've also noticed that the GPU video decode graph is at 0% when watching other YouTube videos which are AV1 or VP9 encoded, I'm guessing the GPU doesn't have dedicated hardware for those.
Does it occur on other sites or only YouTube? Vimeo?
I just watched a Twitch stream for 30mins which was AVC1 and nothing, no glitches. It's all very odd.
I also tried to get the glitches to occur on the other YouTube video that I had found but couldn't get it to occur again, so outside of this one single video the issue basically doesn't happen.
I've also noticed another oddity, if I'm watching the video which causes the fault at 2x playback speed the fault wont occur either.
Unfortunately the upstream bug tracker for Chromium is very tech-heavy and requires a Google account but try to search if you can:
https://bugs.chromium.org/p/chromium/issues/listI checked it out and I think I'll hold off filing a report there, they seem to have no shortage of tickets already, and I'd prefer to nail down clean reproduction steps if I can before making a ticket.
I'll report back if I can figure out anything more concrete.
Thanks again for all your help!
@user1vivaldi said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
I also have an AMD Radeon RX 580, however my current driver version is the latest one from Windows update (31.0.12027.9001, which is from late March) and I haven't experienced any artifacts or other graphical issues after the update to 6.1.
Thanks for pitching in, did you try the video I linked in the first post and hit the seek bar at 1min30, give it a couple of tries it doesn't always occur first attempt?
-
@Pulseammo said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
tested each of the flags in isolation. When rasterization was disabled hardware decoding still occurred according to the task manager graphs (there's one for GPU usage and a separate one for video decode).
Yeah, your issue definitely looks like rasters to me. For those of us old enough, this is what the idea of a "raster" looks like:
https://www.youtube.com/watch?v=PyGj7ekkNqs
So if you can run with only GPU rasterization disabled, you won't get such a performance hit as with HW accel. completely off. Then again, if it only happens on a single video... might not even be worth it
-
@Pulseammo Just tried doing what you said, still don't see any artifacts.
-
@user1vivaldi Why do you guys have such different driver versions?
You say you have
31.0.12027.9001
But the latest ones from AMD seem to be23.5.2Maybe something similar to what MS does with Nvidia drivers? But I wouldn't expect the version number to be so different.
The usual recommendation is to avoid the MS provided drivers (at least for NVidia) and always use the official ones. But maybe in this case the MS drivers do not have this issue?
-
@Pathduck said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
@Pulseammo said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
Yeah, your issue definitely looks like rasters to me. For those of us old enough, this is what the idea of a "raster" looks like:A couple of my friends had a C64 growing up, but I had a NES, so not quite before my time :P.
So if you can run with only GPU rasterization disabled, you won't get such a performance hit as with HW accel. completely off. Then again, if it only happens on a single video... might not even be worth it
I'm just going to leave it on and see if it happens anywhere else, it's really not a big issue as it stands, I'm more curious whether my GPU is on the way out!
Actually I did try some other videos on the channel that I linked from around the same time as the video that causes the issue and the other AVC1 encoded videos also cause glitches there.
Of course I then went to hunt for AVC1 encoded videos on other channels and none of them set off the issue that I could find.
@Pathduck said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
@user1vivaldi Why do you guys have such different driver versions?
AMD drivers have multiple version IDs, the adrenaline package I'm using is marked as 23.4.3 (I rolled back from 23.5.2 which is the latest but both cause issues). But the MS Display driver version in the 23.4.3 adrenaline bundle is "AMD Windows Driver Version 31.0.14051.5006".
Adrenaline is basically what you'd get if you installed Geforce Experience, it's got the extra tools in it for overclocking, screen recording, auto update, etc.
@user1vivaldi said in YouTube Artefacts - Vivaldi 6.1 - AMD Chrome Optimisations [Bug Report]:
@Pulseammo Just tried doing what you said, still don't see any artifacts.
Thanks for trying that for me! I think your driver might equate to 23.3.2 so I'll test that and report back if it resolves anything.
-
@Pathduck The reason for the different numbers might be the fact that the MS provided drivers only include the driver itself without all the additional software like the AMD control panel, recording software, etc. The one that you download from AMD website on the other hand, is an all in one package that includes the driver + the additional user software.
It's not correct to think that only the one from AMD website is the official one because technically it's the same driver that is developed by AMD. The only difference between the two is the method of distribution.
The reason I'm using the Windows update / MS provided driver is because I'm fed up with Windows automatically replacing the AMD provided driver with the one from Windows update, so I gave up and just use the one from Windows update. If you rely on Windows update, on one hand you don't get driver updates as frequently as with the AMD provided driver, but on the other hand, I like the fact that updates with the Windows update driver are very smooth (you don't have to deal with stuff like DDU or factory reset in order to ensure a smooth driver update).
