Linux @lemmy.ml aksdb @lemmy.world 1y ago

AMD GPUs are cursed for me

Each time I try AMD graphics, something is fucked for me. Back with fglrx, fglrx just sucked, so I used Nvidia. Then I had an AMD right around when they finally had opensource drivers, but it was still buggy as hell. So I went with Nvidia again (first a GTX 790, then a GTX 1060). In the meantime I had a new work notebook where I also went with an AMD APU, and had driver crashes for a long time when I was in video calls and it had to decode multiple streams. That thankfully stabilized with Linux 6.4.

Since sooo many people in the community swear by AMD, I thought "dammit, let's try it again for my new desktop" and got an 7800rx ... and I have to reboot ~5 times until I finally make it to a running xserver or wayland session. Apparently I am hit by this problem (at least I hope so). But that doesn't even read nice ... the fix seems to be to revert another fix for powermanagement. So I either have a mostly non-booting card or suboptimal power management.

I start to regret having chosen AMD .... again :-/ I seem to be cursed.

114

114 comments

And here I am with a 3090 having more issues than I have time for wishing I went with an AMD card. Sadly we both can see grass ain't necessarily greener.
- Thanks for that perspective. At least that makes me regret my decision less.
  
  I've tried the open source drivers, the proprietary dkms variant, and standard proprietary drivers and all give me issues.
This reads like an alternate reality for me. I bought a new 3060 ti and using wayland with it is nearly impossible for me. I tried in ubuntu and had tons of errors and in debian/kde it wont even login without x11 enabled.

When you go to protondb.com every game has tons of fixes for nvidia cards and every forum has fixes for nvidia cards while amd mostly works oob.
- Try out COSMIC with the NVIDIA 550 beta driver.
  
  No. I‘m not being dictated by my gpu what OS I can use.
  
  Nvidia is just shit.
Pretty sure the 7000 series is known to be not well supported yet since they're new and didn't have massive uptake, so I don't want to be that guy but...

Some research before hand on what GPU to get from AMD wouldn't hurt?

I've got a 6800XT and had absolutely 0 issues since I got it about a year ago. I see from your replies you're on Arch, so I guess just wait for things to improve unfortunately.
- As I said... I had a lot of trouble in the past and went with nvidia most of the time. It wasn't just a quick shot picking that specific AMD card. My research ended up looking positive. The 6000 series wouldn't have cut it, since the AV1 encoder isn't good enough (or maybe not present at all; I forgot). I also buy this thing to last a few years, so having to take a card from the last generation would have certainly be the point where I just picked nvidia again.
- I’ve been running a 7900XTX for months without issue. Only thing that was missing was some stuff around power setting, fan curve etc but even that I think has been fixed in recent kernels.
Since people normally only report on negative experiences: I was lucky enough to get a reference AMD 6900 XT during the GPU shortages.

Switched from Ubuntu to Fedora for it because Ubuntu didn't have firmware for it yet.

Ever since then it has been a rock solid GPU. Never even had such a stable GPU under Windows.

Have been running Fedora with Wayland for more than 2 years now and can count the crashes on a single hand, most were my fault.

I'm sure once that issue is sorted out that GPU is going to ride along for years with minimal maintenance required.

(You might want to downgrade your kernel until then though)
- I couldn't get my 6900XT to drive my G9 at 240Hz, but 120 isn't too bad. I should probably try again soon.
  
  Been 20+yrs of some random flavor of driver problems for me, since my 9700 Pro at the very least.
  
  Over DisplayPort? That's interesting, I knew AMD can't do HDMI 2.0 but there shouldn't be a problem with DP.
  
  Might wanna try a proper new certified DP 2.1 cable, just to be safe.
  
  I "only" drive a AW3423DW but no issues at 3440x1440 with 165Hz.
Yup I'm hit by the exact same bug currently. But I was able to go back to before I updated with Snapper and now I'll wait until the fix is in the Tumbleweed repos.

But other than that I'm much happier with the AMD than with my Nvidia (on Linux that is). VRR with Wayland on multiple monitors just works without issues. And before this week I never had any issues at all with the 7800XT.
- I need to give the LTS kernel a shot tomorrow, but I could swear I tried that and had the same issue. Which now makes me fear that I might have a different problem. Argh.
  
  Dammit, same symptoms. Which, I guess, is not a good sign. Maybe my issue is different or I have another issue on top.
What kernel version are you using? 6.7? Unfortunately using the latest and greatest kernel means you'll be among the first to get bitten by new bugs. Does the issue also occurs on 6.6 and 6.1?
- Same issue on 6.6.14. :-(
Funnily, I only run AMD now for the same reasons, except with Nvidia as the PITA. Always ongoing driver issues, power management or fans running like jet turbines.... Last 3 machines AMD, no issues with the GPU's/drivers.
On EndeavorOS I haven’t had issues with a Vega64 and now with a 6800XT. I followed the AMD Gpu guides from Arch wiki to get everything up and running but that was back when I started the build with the Vega 64. After the upgrade I didn’t even need to touch anything and all non anti-cheat games work quite well. Maybe I got lucky though.
Sorry to hear that. For what it's worth, I've had no problems with integrated AMD graphics, so maybe it's a PCIe issue?
- Hmm, interesting idea. I need to investigate that. The dmesg output is full of amdgpu irq errors, but of course that could also happen with an issue on the board.
  
  I would rule out a generic hardware issue, since 1) I get graphics during boot up until it needs to do a modeswitch (I guess) and b) it works fine so far on Windows.
  
  I did have a similar issue after the first boot on Windows as well and assumed so far that the modeswitch after the initial driver install caused the problem. But Windows likely also installed chipset drivers at that time, so PCIe could be a possibility. Then again... I know that Windows reloads graphics drivers on-the-fly... but chipset drivers? Probably not. Which would speak against that theory.
  
  I have no clue how Linux initiates the communication with a PCIe board, and whether the amdgpu driver would take care of that. But hardware excluded, some misconfiguration on the driver's part could be present. Good luck!
I just got a 7600XT. My only complaint is that it isn’t pushing quite enough frames so I would need something more beefy, but then I will also lose GSync because of my monitors so I will probably simply return it and go back to the 3080. Lower TDP and thermals was quite nice though and wayland was much less buggy. No crashes, I’m on ubuntu tho.
- The 3080 has ass performance on Wayland
  
  My favorite bug is when I resume from suspend and everything becomes rainbow colored.
I had a rock solid AMD RX 580 up until the release of kernel version 6.7. Now I'm lucky to get a system that can remain up for longer than thirty minutes. Sticking to 6.6 has worked for me and definitely something you should try as well, but it's worth noting that any amount of time spent on the issue tracker for AMD GPU stuff will reveal tons of issues from 6.6 as well.
I have a similar story with an RX580, I replaced my GTX 1060 3GB for a 8GB RX 580 mostly because the 3GB of vram were an issue for BeamNG.

Now I can't record my 3 displays with the RX 580, it just fails when trying to do so, and 2 displays results in constant encoder overloads, something that the 1060 had issues at all, also my colors are off when recording and I have no idea why, it even happens when recording with the CPU:

https://bbs.archlinux.org/viewtopic.php?id=292196

Also kernel 6.6 broke the power reporting on all polaris GPUs, thankfully that was fixed recently in kernel 6.7.2, but holy shit it took like 6 months to fix that.
- I probably shouldn't have read tests and forums, but simply searched for crashes and open bugs to get a feeling for what I am getting into. Then again I also read from people with very ugly problems with nvidia, so it's not a really good measure.
  
  I really want AMD to be good; they offer more VRAM where nvidia always seems to cheap out in pretty suspicious ways. Then again nvidia seems to be more power efficient.
  
  My time with nvidia on linux was 0 issues in performance or usability.
  
  The only sort of issue that I had was that the GTX 1060 drew 20W at idle when using the 3 displays, this was a bug that nvidia fixed for the RTX 20 series and newer cards but never fixed for pascal lol.
  
  But even on BeamNG, there was a period were the native linux version didn't work on mesa while it worked for nvidia, now to be fair with amd this was because the vulkan implementation of beamng is horrible and right now it does not work on either lol.
- Polaris GPUs had very weak video encoders, I also had an RX 580 and had issues on Linux as well as Windows. To my knowledge the AMF encoders worked better for those, but I could never get them working with OBS
  
  Oh I did try to use the AMF driver, my first attempt ended with i3 crashing upon startup. What was worse is that even after removing those drivers and putting mesa back it still crashed on startup, good thing I had a btrfs snapshot before messing with that.
  
  My second attempt I was able to use the AMF on OBS, but it still failed to record the 3 displays.
  
  My biggest issue right now is the issue with the colors, I don't care if I have to use the cpu to record at this point.
It could be your monitor or even monitor cable. I have this monitor which absolutely fucking refuses to work with AMD oved HDMI. If you have inexplicable system sleep issues, black screen issues, startup issues, etc. It could be the monitor at fault
- Thanks for the suggestion!
  
  While it's a possibility, I think it's unlikely, since the machine works fine with Windows. I also compiled the tkg 6.7.2 kernel which includes the revert-patch for the offending change and so far the machine booted three times without issues, so it seems to fit.
  
  That doesn't rule out the possibility of display issues tho, back when I had the faulty monitor it was much more severe under Linux, I never managed to track it down tho (using AMD hardware for over 10 years now, this one issue busted my nuts pretty hard)
  
  If you have a TV or something, at least try it to rule out possible outside factors
I've always had great experiences with AMD and not Nvidia. Maybe its just there newer cards.
RX 6700 XT here... once I refreshed the thermal pads and the thermal paste, it works great in Windows and Linuxes... Ubuntu, Mint, Fedora, Bazzite (Immutable Fedora but for gaming), it had no issues with the amdgpu driver builtin on any of them.
- It's a completely new card, so I will not fiddle around with it. Also it runs almost flawless on Windows (aside from a similar crash on the very first boot during driver install).
oh man, reading the comments fill me with fear, as I just ordered a new computer after stretching my old laptop for 8 years or so. I was super close to getting an AMD but went with Nvidia in the end... but so much bad juju in the comments for Nvidia too...
- You may wish to pick a distro that makes a point of nvidia compatibility.
  
  I use nobara, who have a few options in the welcome script specifically to improve compatibility with nvidia. I've specifically heard popOS mentioned several times as one people have liked with nvidia as well.
  
  Some only ship with or distribute alternative open source nvidia drivers that tank performance.
  
  Thanks for the advice, but a distro change for me would be a huge annoyance. I haven't have issues with my laptop's 1060 nvidia on Arch, and never had issues with the proprietary driver.
  
  My worry is that even though mature GPU are probably well supported, I bought a relatively new one (4070 super ti) so maybe the new models have some issues due to having more features/being more extreme. Most complains here are about 30/40 models after all.
- Yeah, until this thread I was convinced I should stay away from nvidia GPUs when building a new PC with Linux in mind, but I'm not so sure anymore.
I use an AMD 7900rx with an AMD 7950x processor since almost a year with Gnome / Wayland on Arch. No problems up to now. Yes, I am a gamer too.

As others said it depends on the distribution you use.
- Arch is not exactly homogeneous. Which Kernel package and version do you use? Which firmware package and version?
  
  I use Arch, btw, and have these issues with default kernel 6.7.2, 6.7.3 and lts kernel 6.6.14. Firmware package seems to be from 2024-01-15, IIRC.
Me playing with my 7900XTX no issues on Bazzite

https://github.com/ublue-os/bazzite/#about--features

https://github.com/ublue-os/bazzite/#features-from-upstream
It also matters what Linux distro you have. Some of them are horrible. I'm super happy with amd graphics on arch, and have no issues whatsoever, with probably 30 games in steam library that all works very well.

So I think it may be your system and what drivers you installed, or some other config.

I have a 6900 XT card, latest kernel, latest drivers. But I've had this graphics card since kernel 5.8 I think, with no issues.
- I am running Arch here as well. Most people who referenced that issue I linked also seem to come from Arch. So it seems like a problem due to the "latest" kernel. I don't accept this as an excuse, though. It's still a stable kernel an I don't expect drivers to be published that were not tested in advance. And it looks like this has happened here. Maybe bad timing on my part and this was/is the only hiccup in a long time (see "cursed"), but I guess I'll find out.
  
  Having a bleeding edge kernel can and will come back to bite you. There's a reason why many distros hold back with kernel updates for so long, there's issues that only can be found with user feedback.
  
  From experience, "stable" in the kernel world doesn't mean much unfortunately. I encountered dozens of issues over various versions and different hardware already and it's the main reason I don't run rolling release distros on my main rig.
  
  There's also been enough times where the latest Nvidia driver borked my test system at work so I'm fine with just not running the latest kernel instead.
Me with a Vega 64... the forgotten platform. A few games will just straight up reset my gpu with certain instructions, taking the whole system with it. I can't even play Minecraft with a Mesa version newer than 2 years anymore due to regressions.

Good thing to know 7800 XT is also cursed though, I was planning on getting that one to escape my situation. lol.
- Kinda weird, is the first gen Vega Apu different enough to not have these problems? Cause I've been pushing that thing hard enough it's starting to have actual hardware faults, very rarely had software related crashes that couldn't be resolved with a temporary kernal rollback
- My PC mysteriously reset itself every now and then, and I got Vega 64. Is it because of that?
Run sudo dmesg | grep amdgpu and look for errors.

You may have a firmware file missing, for instance. If that’s the case, it’s an easy fix - just download the firmware files from the kernel tree and put them wherever your system wants them.

This is how I do it on Debian but it should be easy enough to adapt to whatever distribution you’re using (it might be exactly the same tbh): https://blog.c10l.cc/09122023-debian-gaming#firmware
- Thanks for the idea!
  
  dmesg shows the same errors as in the referenced bug ticket. So I don't think missing firmware is the issue. I would not be surprised however, if the problem in general is a combination of amdgpu and firmware behavior. (IMO the hardware should not crash as hard as it does, so the firmware seems to be a bit wonky too)
KDE Neon user here, I have not touched any graphics settings and my AMD card runs 32:9 120 Hz flawlessly
the most bug-free gpu experience I have with Linux is Nvidia GPU + KDE X11 with compositor disabled. Pure bliss. I've had a 6700XT and it was terrible too, now I have a 4070. For my laptops, intel igpu works decently well with wayland KDE, but there are few bugs, like having to clear some apps gpucache (vscode) quite often
- At least with my 1060 compositing wasn't an issue. But true, I rarely used Wayland. Do you have specific issues when compositing is enabled or do you just prefer the simpler rendering?
  
  I prefer without for the aesthetics but also for functionality: compositing x11 with multi monitors of different refresh rates is still broken, everything becomes locked at 60hz instead of the max for each monitor.
Using amd GX 6600... Mostly going fine, tho I haven't tried any big heavy games. One thing tho... Everytime I turn on my computer, no display. I reboot it and then ot works fine, but ot never does the first time. One path I'll investigate is the monitor: my monitors are both older and use DVI or VGA ports, so I have to use converters. I might try and get my hand on a more recent monitor to see if I still get the same problem. But if I do, I'm not even sure where to ask. I don't even think it's a linux problem, because I tried removing my drive with linux living one with windows and the problem remains. I also was using mint when the problem started and switched to Arch (btw) since and it doesn't change a thing.
- I had a similar problem which was resolved by disabling the motherboard integrated graphics in bios settings.
  
  Thank you ! It didn't seem to work on it's own, but I also noticed I wasn't booting in EFI mode, so maybe if I just change my booting partition and combine it with your advice it'll work...
I've had similar issues. I don't understand the love for AMD. My whole rig is AMD, but it's constantly having GPU crashes. All games run at high FPS and my CPU temps seem nominal. But the games will crash. Everything from RimWorld to Baldurs Gate 3. They all run pinned at 60fps but randomly crash. I've tried a thousand different configurations and drivers. I've tried Ubuntu and Linux Mint. I'm now just accepting that I can't rely on it as a gaming rig. I like that AMD is trying to be progressive with open source drivers but the quality doesn't seem to be there. My next rig might be Nvidia and Intel. But we will see.
- Weird. I would check PSU next.
  
  My issue was the GPU fan and the PSU fan would blow into each other. I opened the PSU and reversed the fan
  
  Maybe 🤔
- It runs Wayland. That's all the reason I need.
  
  What does that mean? Genuinely don't know what it means that it runs Wayland.
- Did you check the system logs to see what caused it?
  
  Many things can result in seemingliy random crashes. Any overclock (including XMP and Expo) or undervolt or even a bios version can be problematic.
  
  I would check first if it's stable on windows.
  
  It's not stable on Windows either. But I haven't looked at logs because I didn't really know what - or how - to check.
Update: today I was able to update to kernel 6.7.5 and the issue disappeared for me.
Ohh, so that's the bug I've been experiencing ever since Fedora 39 updated to kernel 6.7. But I only get this on restarts, so cold starts work just fine. I actually have a 7800 XT as well.

But other than that I only noticed one issue: video playback in Firefox sometimes shows visual artifacts across the screen while a game is running in the background (well, with Baldur's Gate 3 at least). Fedora 39, KDE Plasma. Kernel 6.6 or 6.7 (or 6.5 for that matter). That said I also had some suboptimal experiences with browser video playback on an AMD APU notebook under Windows (severe framedrops), so I'm not sure where to point my fingers at.

Other than that it's honestly been great. I switched from Windows + Nvidia to Linux + AMD basically January 1st of this year and only ever booted Windows twice to transfer game saves over for the few games that don't have Steam Cloud.

Turns out most of the problems I had with Linux desktop was with Nvidia. I spent more time troubleshooting than actually using software. AMD isn't perfect on Linux and with new kernel versions you're suspect to run into more issues, but AMD (and Intel) mostly work out of the box.
If it makes you feel any better you're not the only one. I also have this problem. Whenever it was time to upgrade my video card I'd try Ati and later AMD and it would always have some annoying issues. Meanwhile I'm on my 7th or 8th Nvidia card over the years and they're always great.
I don't know if you've tried it yet, but having recently installed 6.7.3 I noticed a whole lot of amdgpu fixes in the changelog. Maybe it will help?
"I keep hitting myself in the foot with a hammer! Why does the hammer keep doing this to me!"
- If other people apparently have no probiems, it can't be AMD in general. Never trying AMD because of a bad experience 10 years ago would also be extremely unwise. "We've always done it like this around here" is how you fall behind the times.
  
  What distro are you even using?
Linux is cursed for me

Sorted that for you
- Not really. I use Linux as my main driver since about 2006. My intel laptops and the mentioned nvidia gpus work fine.

114 comments