PC Graphics

[tapekeep]

This happens on every browser I've tried: I load a 4K60 video on YouTube and there are dropped frames on Firefox and very sluggish scrolling on Edge (likely the same on other Chromium-based browsers), the video even stops to "buffer" when scrolling the page with the video window in view. This happens with every Adrenalin version I've installed up to the most recent WHQL version to date (22.6.1, Windows 11).

An easy solution on Firefox is to go to about:config, search for DXVA and set "media.wmf.dxva.d3d11.enabled" to false. This removes the stuttering and makes everything smooth, as it should be, but CPU usage goes up quite a bit.

Another thing to note is this issue does not happen with the drivers supplied by Windows Update, though they are very out of date. This issue also doesn't happen at all in the Linux distributions I've tested in the same system, so it seems to be exclusive to recent Windows drivers.

The system: Ryzen 9 5900X, 64GB DDR4 3600MHz (4x16GB; this also happened with 2x16GB at 3200MHz), ASUS TUF Gaming B550M-Plus (Wi-Fi), PowerColor Radeon RX 6900 XT Red Devil, Windows 11 (also tested on Windows 10). Tested with PCIe Gen 3 and Gen 4 modes, ReBAR enabled and disabled as well as Gen 3 and Gen 4 NVMe SSDs as boot drives. I had a Vega 56 in this system prior to this card and none of the previously described issues happened at all.

[westclox]

@tanazzz - can you provide a link to the thread? 
The issue is back for me as well, so I've had to reenable the d3d9 workaround. This is ridiculous.

[tanazzz]

There's nothing of note there. No useful responses and nothing that we've not discussed here before. Only people saying whether they have the problem or not. I don't know who we can turn to. The driver team do not care and there's no way to contact them.

[tanazzz]

I keep reporting the bug in Wattman every single driver version, tried DMing AMD Radeon's X account with a link to this thread. Still nothing.

[tanazzz]

23.8.1, clean driver install using DDU under safe mode, latest Win 11 build - bug still persists. Worse on Firefox than in Chromium based browsers but still present in all browsers.

[tanazzz]

23.8.2, again clean driver install and latest Windows 11 build - bug still persists. I'm not giving up on this, gonna keep bumping the thread every driver version since there is no proper bug reporting process and this still has not been addressed more than a year later.

[tapekeep]

I'm not being thorough with driver cleaning, but it's all the same here. I think the thread is just being ignored at this point.

[tanazzz]

Oh I know there's probably less than 10% chance of the issue getting fixed and that % is getting lower the older RDNA 2 gets but I just want to have  a documented thread with hundreds of comments demonstrating AMD's ignorance and incompetence the next time someone asks me if Radeon drivers are buggy.

[DimkaTsv]

I only half understand what are you talking about. 

Yes, 4k60 playback in Firefox YouTube feels choppy, i can confirm that part. But i have no frames dropped, neither i have same issue on Chrome or Edge. On contrary, playback in Chrome is smooth. Scrolling in Chrome and Edge is also fine, but there is tiny jitter one time when video hides due to scrolling past it. 6750XT and 23.8.2 btw

Are you sure it is AMD problem and not Firefox's? Because stuff works. 

And also... DO NOT TRY 8k60. RDNA2 can only do 8k30 decoding. If video falls up to 8k60 from 4k60 and starts lagging horribly - it's not driver issue either. 

[tanazzz]

I have the 4k video playback choppiness on all browsers, even tried Chrome on Ubuntu and playback still wasn't right.

[DimkaTsv]

If even linux is not normal, then something is definitely wrong. What can it be though... Tbh, no idea. I do remember quite old driver versions that had terrible frame skips on Chrome. Like REALLY bad ones as about third of frames on 4k60 were gone. 

Had you tried open source drivers for Linux? They are not from AMD, so maybe something will be different? Can it be something with monitor and refresh rate adaptation to playback framerate (which should not happen in theory, but who knows...). Can it be some random ass load stuttering system? (there are quite a lot of those... But on linux such is unlikely). 

What version of W11 do you use? AMD doesn't quite fond of Canary one. 

But i assure you with all issues i encountered, jitter in Chrome at 4k60 is not one of them. At least definitely not at current point of time. 

But, right now i spent a day to find issue that occasionally broke signal to monitor through FreeSync variable vertical blanking. Tbh it is combined AMD and monitor manufacturer issue, but still... Tried to resolve 3 different issues at once:

1. VRAM being maxed out at idle with 30W PPT

2. FreeSync works to solve issue 1, but then occasionally VRAM went to 0 mHz mode and caused different issues like image freezing for 2 seconds after moving to games, potential crashing due to recording and moving into game

3.  Local image corruption or flickering during alt-tab or hovering on top of some elements on webpages in Chrome. 

Well, anyways, it does work, but then it can randomly cause data stream to the monitor breaking. Well, I thought it was something with resolution setup, but no. It looks like monitor issue. Edited lower range of FreeSync from 48 to 50 (49 also works, but round number...) and everything should be fine now based on quick tests. But i doubt this one is AMD fault and not monitor manufacturer one. Well, not that anything changed for me as range below 54 FPS was covered by low framerate compensation specifically (causing it to have doubled refresh rate), as this ranged still works as it did. Literally no idea what was broken. 

[tanazzz]

The open source driver IS the driver that I've tested it with in Linux because that's the driver most people use by default. I'm almost 100% certain it's an AMD thing because older drivers 22.5.1 and below don't have that issue. As far as monitors go my setup is quite complicated. I have 4 monitors, 3440x1440 100hz, 2560x1440 240hz, 2560x1440 165hz and a 4k120hz TV. All support free-sync but I have it disabled on all displays as otherwise I have occasional black screen issues and driver hangs.

[EnderW12]

Firefox
about:config
media.hardware-video-decoding.force-enabled
true


Fixes the problem completely for Firefox on Linux (ubuntu 22.04).
No solution for Chrome yet.

[tapekeep]

IT GETS WORSE!

23.9.3 doesn't let me have two videos playing at once without grinding the ENTIRE COMPUTER to a halt! Doesn't matter if it's Firefox, VLC or MPV, I have to Alt + F4 while the whole desktop runs at less than 1fps. The D3D11 DXVA also has the same problem as before, lags on 4K60 videos.

[tapekeep]

Update: a reboot fixed the "computer grinding to a halt" issue at least temporarily, but the other one still persists.

[tapekeep]

23.10.1 was apparently pulled, but I managed to install it. Unfortunately, the problem still persists.

[tapekeep]

23.11.1: it's worse now, even with the DXVA trick the video will freeze for two seconds before catching up. Without the DXVA trick, the framerate takes a massive hit, much bigger than previous versions.

[tapekeep]

I just bought an ROG Ally, which has an RDNA 3 GPU and Windows 11. I installed Firefox and, sure enough, the issue is still there and the DXVA bodge makes it disappear. I'm on the latest drivers supplied by ASUS (23.10.2).

[warchild]

How this has not been resolved yet, after such a long time is nothing short of a disgrace.

AMD, please for the love of god find a solution for this 2 year old problem, rendering watching 4K material a sluggish unbearable experience. This is 100% costing you customers. It's in fact so bad that I myself would question staying with AMD for any future purchases.

The workaround proposed here does "work", however has a significant impact on CPU load (~15% on a 5900X).

I found that the following workaround seem to resolve it, with minimal performance losses:

media.wmf.zero-copy-nv12-textures-force-enabled to true
gfx.direct3d11.reuse-decoder-device-force-enabled to true

I understand that Firefox might not have the majority of the market share, but you should expect the most high-end card from AMD to be able to play 4K material on the biggest video platform without slowing down to a crawl and have no where near fluid output.

[tapekeep]

Tested and working on latest drivers (23.12.1), thank you. The other bodge also works, but default settings still lag all the same.

[PwnedHusky]

Thank you! This is a good workaround for firefox. I've been experiencing this issue for so long and people keep pointing to suggest to disable hardware decoding which makes my CPU goes at 90%.

[thedreadedgman]

I've been watching all the bugs I could find in bugzilla and now here in these forums... it's been years with CPU-decoding... I can't believe these two settings actually worked!? wow

[joshthornton]

My 6900 xt bit the bust. Sad day. Got a 3070 ti dirt cheap until the 8000 series is out. Was watching 8k videos with 0 frames skipped. Forgot what that was like. 

Another funny thing is that my integrated graphics in my 7950x3d seems to work much better than my 6900 xt ever did on YouTube. Go figure.

It's pretty sad that this is still going on. It's youtube. **bleep**ing youtube. My intel laptop with integrated plays back videos better. How is this acceptable?

End of rant. I feel for you guys. Here's hoping it gets miraculously fixed.

[westclox]

It's not acceptable. I think this goes to show how amazing corporations are at listening to user feedback, and also at proper QA testing.

If there are a few people complaining here, this probably means that we are a more or less representative sample of many more users (tens if not hundreds of thousands, probably) affected by this bug.

Do they care, though? Clearly, not really. 

[tapekeep]

24.1.1: issue still persists.

[sebarobbie]

Till this day i have been experiencing the same issue as well. Currently on 23.12.1 since the latest drivers are stuttery and the only solution is to disable "media.wmf.dxva.d3d11.enabled". In my case , disabling this feature makes firefox somewhat slower and its not the ideal solution. Hope we get a fix some day.

RX 6800 XT. Tested on Windows 10 and 11.

[CitRus2010]

I have the same issue on a Radeon 6700XT with the latest drivers 24.2.1. I tried playing AV1 4K videos in Firefox stable 123 and Firefox nightly 125 on Windows 10. At 1440p, there are slight stutters, at 4k they become very noticeable, and at 8k the browser completely freezes. When I disable hardware decoding and process with the CPU instead, this problem does not occur.
The situation is a bit better in Chrome, but there are still lags.

[CitRus2010]

I submitted a bug report on bugzilla. Idk maybe it needs some upvotes https://bugzilla.mozilla.org/show_bug.cgi?id=1884529

[tapekeep]

24.3.1: no, I didn't forget about this, it's still broken. I'll do another fresh Windows install on a brand new SSD, don't think it will do any good but I'll report back anyway.

[tapekeep]

Fresh Windows install, factory driver reset: issue still persists.

[tanazzz]

Can confirm as well that issue is still present in 24.3.1.  Wondering if anyone from AMD is even reading this or at least the many bug reports sent through their driver. Highly doubt it.

[westclox]

Why would you think proper video reproduction should even be a priority? That's ridiculous. Please adjust your expectations, buddy -- this is only a multi-billion dollar corporation. 

[tapekeep]

24.4.1: no change, I still have to do the about:config bodge to get things working. One not so amusing fact is that the GPU usage in Task Manager jumps to about 95% whenever a 4K60 video is playing on YouTube. It's a 6900 XT, for crying out loud!

[tanazzz]

Same. Getting rid of my POS 6950XT monday when my 4080 Super comes. A "bad" purchase decision on paper but I've spend almost 2 years dealing with this and all sorts of other crap form AMD. Good riddance team red GPU team, hope we never meet again.

[tapekeep]

24.5.1: It seems ever so slightly improved, but still laggy and unwatchable.

[tapekeep]

24.7.1: I'm still bothering to update this thread despite having lost all hope of this bug being fixed. It isn't fixed.

[EnderW12]

I had to use the iGPU HDMI out of my Ryzen 7 5750G - this makes the Vega be default video decoding device so i can watch video and listen to anything via Bluetooth hadpones which were disconnected all the time while i was using my RX 6800 XT (I assume it's due to completely broken audio pipeline).
Sideeffects also include some benefits: dGPU is always at lowest power state consuming 6W and the Vega iGPU is times more efficient during desktop use and video playback.
Gaming performance loss is barely noticeable for me.. and I undervolt my RX 6800 XT anyways
I am using Linux (PopOS 22.04) - perfect experience with low energy bill for me.

[dancfuller]

Thanks for continuing to upgrade drivers and update this thread. I really appreciate it. I don't love updating graphics drivers unless there's a benefit to a game I'm playing (or it fixes a bug... like this one), so I always check for your posts after a new driver is posted.

[DimkaTsv]

And yet i only encountered this issue about 2 years ago with late 22.X.X drivers (gosh i don't even remember which specifically, so long ago it was. But it basically dropped up to 60% frames, which was pretty darn noticeable and reported by debug information), after which AMD fixed it relatively quickly. Since then 4k60 worked perfectly fine in Chrome, Edge and Firefox.

And i am saying that as person who tried to reproduce it on 6750XT and 7800XT... AND I REALLY TRIED! SEVERAL TIMES! Heck, even MPC-HC had played 8k30 video worse than browser due to splitter overhead (8k30 on 6750XT, or 8k60 for 7800XT. And other players work fine, just MPC-HC had this overhead).

I literally just reinstalled Firefox on recently reinstalled video, loaded some 4k60 sample on YouTube and it played smoothly. I also did same with 8k60 with same result, no problems (granted, i am now on 7800XT). 

So... Maybe issue is with your particular HW/SW setup, and not AMD drivers? Just suggesting... Had you ever done more specialized tests? Like offline video playback (of same exact video, which you can download from YouTube) and not in browser YouTube, or some decoder/encoder benchmark like ffmpeg or transcoding application? Do you have particular video you do your tests on, or just random one that  fits "4k60" definition? Your issue description is too shallow, and provides barely a surface level of useful information. As if with all this time you hadn't EVER tried to look deeper into cause. 

Anyways, what software fix from AMD do you expect, if issue is: 

1. Not widespread, meaning it won't have high priority (that also means that it can take YEARS before they reach this bug in tracking list. In stuff like ffmpeg there are few dozens of bug reports daily, and it is not a driver)

2. Not easily reproducible due to unknown variable? (Aka why wasting your time as developer when you can fix more relevant stuff?)

Do you expect a fix on level "randomly pointing a finger at stars and naming them, until you guess one correctly"? 

Unless you know why or when specifically it happens (in sense of "under what specific conditions"), do not expect quick fix, because noone will prioritize such issue over ones that are easy to test/fix or critical by nature. 

P.S. Also based on your older messages your GPU load getting to 95% is both kinda expected and weird.

Kinda expected because HW decoder is not core, it doesn't care about what GPU model you have, it's physical size is fixed and same on every GPU from same generation! There can be some exceptions, like professional GPU's and higher tier models having more encoder blocks allowing for higher encode performance. (For example on RDNA2 you will have 1 encoder on low tier GPU, and 2 encoders starting from 6800XT. On RDNA3 you get 2 encoder blocks starting from 7700XT. But decoder is still a single unit on every consumer GPU.). Meaning load on decoder will depend only on video codec/resolution/framerate (and slightly on bitrate and other stuff). 

And it also weird, because RDNA2 HW decoder supposed to be able to decode up to 8k30 video (even though with pretty small overhead, something like 34 or 36 FPS is limit 8k60 is too much). But still, it should've given you only about 60-70% codec usage by report. Maybe spikes on bufferisation) Why is it at 95% on 4k60 for you? What else eats into decoder?  

And if it is CORE which ramps to 95%, then i have even more questions, because HW decode literally isn't done via GPU core. Meaning that if it is SOMEHOW done through core, then something went REALLY wrong, and you must seek where you as user f*cked up (i don't even think that it is possible for HW decode to fallback on GPU core instead of CPU decode). 

Try to find answer to this question, and not mindlessly send messages in thread.

P.P.S. If it doesn't happen with old driver (aka 22.5.1), then it is probably something related to DXNAVI libraries, and should've been "fixable" (or at least "testable" for causation) by changing registry up until... Start of 2024? Frankly i don't remember when they removed DXNAVI from drivers... Oh, wait, i think it was when they split drivers into RDNA+ and pre-RDNA branches. But still, why tf this problem doesn't affect most users, but affects you? 

P.P.P.S. If you wanted to reach AMD developers from forum, you are oblivious. This forum is only for user to user exchange. There are moderators, but developers only look into "for developers" section of forum, which requires approval/invitation. But then, they also have competely different approach to topics as well. Bug Report Tool exists for a reason, and making proper bug reports (and not incoherent mess) is quite tough actually. Issue also must be reported at least by some number of people, as single report is statistically insignificant (aka nobody knows if issue is not just some random HW malfunction or FW/SW conflict and actually is driver problem). 

[tapekeep]

24.8.1: I'll copy and paste warchild's Firefox bodge from a couple posts back. The issue comes back when these are set to false, their default values. I would like, once again, to emphasize that Firefox isn't the only application that suffers from stuttery video performance and I have to resort to software decoding on some legacy programs to avoid this issue.

In about:config, set

media.wmf.zero-copy-nv12-textures-force-enabled to true
gfx.direct3d11.reuse-decoder-device-force-enabled to true

If this doesn't work, set the above back to false and set

media.wmf.dxva.d3d11.enabled to false

[tapekeep]

24.9.1: I'm getting acquainted with how much of a dumpster fire the RDNA2 video encoders are, trying to get a usable H.265 3840x2160 59.94fps capture of a VRChat session. No luck thus far. Anyway, back to the topic: no, it hasn't been fixed, bodges above are still required. I seriously hope RDNA4 brings media encoding and decoding back to parity with the competition (not that I'll be able to upgrade anytime soon).