You are not logged in.
Hello!
I have a computer with an AMD Ryzen 7 5700X, a B550 motherboard, and a Gigabyte Radeon RX 7600 XT Gaming OC (more info in signature). I'm running Arch Linux with GNOME, on Wayland. I have the default kernel (6.10.3-arch1-2 at this moment) and the LTS kernel installed. My system is updated. My single monitor is LG 29UM69G-B, 75 Hz, 2560x1080p, FreeSync capable. It is connected using HDMI.
I bought this card some months ago and noticed a strange behavior: if my computer is doing nothing or if I'm doing tasks that require almost no graphical processing - like reading PDFs or editing documents in LibreOffice Writer -, the GPU shows strange usage peaks. I discovered this problem because this GPU from Gigabyte is quite noisy at the moment the fans start running. So I installed Mission Center (Flathub) and let it minimized to record the GPU behavior. The result is this pattern:
On the left, the card should be idling. There are no apps opened (and nothing is running in the background). There are huge 100% usage spikes followed by 0% usage periods. As a result, the card temperature rises above 50ºC and the fans turn on. Then it cools down until the cycle repeats, and it's extremely annoying. The apps tab in Mission Center also shows the high GPU usage, but not related to any app. Even while I'm writing this topic, with this one tab opened in Firefox, the issue happens.
On the right, I started to actively browse the web and watch YouTube videos. The card works as expected and stays cool (rarely going above 40ºC), never turning on the fans, even hours later.
I thought this could be some kind of regression, but it always happened, since I got this card months ago. I did some research and found an issue with the VRAM frequency on AMD cards being locked at 100%, but it was resolved a while ago and the VRAM of my card works as expected. I also tested with the LTS kernel and got the same result. I even reinstalled Arch with KDE Plasma and the result is the same. Back in GNOME, I thought VRR (I use it since my monitor is FreeSync capable) could be the issue, but I disabled it and the result is the same.
Besides this, my system runs normally, and the card plays games perfectly. Any thoughts? I don't even know where I should report this issue.
Thanks in advance.
Last edited by KairanD (2024-08-10 19:13:57)
R7 5700X | CNPS 9900 MAX | B550 Tomahawk | 32GB (2x16GB) T-Force DDR4 3600 MHz | RX 7600 XT Gaming OC 16GB | SSD MP600 Pro XT 2TB | HDD 1TB | MAG A650BN | PCYes! Horse | Arch Linux
Offline
You can use tools like radeontop or amdgpu-pro to monitor GPU usage in more detail:
sudo pacman -S radeontop
radeontop
Offline
although unlikely - have you scanned for any unknown files like a crypto-miner?
I know this sound more like a windows issue with an infected system - but the symptoms are compareable
Offline
You can use tools like radeontop or amdgpu-pro to monitor GPU usage in more detail:
sudo pacman -S radeontop
radeontop
Hello! Thanks for the recommendation. I have tried radeontop and found something interesting. Radeontop cannot detect which GPU model I have, but it works. The crazy part: as long as I keep the terminal running radeontop opened or minimized, the GPU clock stays around 800 MHz all the time and it never heats up! I kept psensor open to check the max temperature and GPU fan activity during a whole day of work and the problem never happened (GPU stayed below 40ºC and the fans never turned on). This behavior is shown in the figure below:
If I close the terminal window with radeontop, the problem comes back: clocks drop below 800 MHz (Mission Center shows numbers as low as 33 MHz), and eventually the card shows peaks "working" with nothing, heats up and the fans turn on, until it stops and repeats.
That's bizarre.
although unlikely - have you scanned for any unknown files like a crypto-miner?
Hi there! Thanks for your reply.
I'm sure my system is clean, without any form of malware. I have even installed Arch again in a 120GB spare SSD I had lying around and the same problem happened with the new install (and a different desktop environment - KDE Plasma).
I opened an issue (https://gitlab.freedesktop.org/drm/amd/-/issues/3549) and another user reported the same thing.
I'm far from being an expert, but I think there may be something wrong going on when the card enters an idle state of low power consumption. Another interesting fact: Mission Center identifies the max TDP as 165W (the same Radeon RX 7600 has), but the RX 7600 XT should have a 190W limit.
It may be a Radeon RX 7600 XT specific issue.
R7 5700X | CNPS 9900 MAX | B550 Tomahawk | 32GB (2x16GB) T-Force DDR4 3600 MHz | RX 7600 XT Gaming OC 16GB | SSD MP600 Pro XT 2TB | HDD 1TB | MAG A650BN | PCYes! Horse | Arch Linux
Offline
as for the max tdp this seems something linux specific but more general
I have a sapphire pulse 7700xt and windows can push it to max 230w while linux is limited to just 200w
so I guess if you would test your card on windows it wiuld go up to the full 190w - and maybe wouldn't cause this issue
I'd like to help further but I cannot reproduce the issue with my gpu
Offline
Moderator Note :
Several posts were split off to dustbin, please keep things civil.
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
clean chroot building not flexible enough ?
Try clean chroot manager by graysky
Offline
<Moderator Note>
It was brought to my attention(thanks, seth) that one of the posts I split off has information that is relevant for the thread.
I have quoted the post fully below.
<End of Moderator Note.>
I thought I was going nuts
0. I have exact same issue (same gpu). Opening up `radeontop` or `amdgpu_top` "solves" the issue. HOWEVER the GPU is stuck at higher clocks than without it, mem iss stuck at 456MHz (vs 96Mhz that it's supposed to be) and GPU being at ~820Mhz (vs 255Mhz that's set to be) -- Check that with LACT
what is MORE interesting is... running vkcube makes GPU clock go ~500MHz, while total idle (with radeontop) is... 800Mhz and higher power draw - 12 vs 15W.The issue is definitely real, it is not a bug in the presented data, because I can see corresponding pattern of energy use at the metering power plug -- it spikes even up to 80W every 1-2s.
Make sure to increase the resolution of your measurements. I'm using `nvtop` and I have a 0-100-0-100% pattern most of the time.
1. Did you figure anything out?
2. The TPD is actually.. correct. Basically the reported TPD in the utilities is ONLY for the GPU chip itself, while total power is the one reported by manufacturers (VRAM, efficiency etc) -- I confirmed that with a power meter, as i use this gpu as an egpu, so only gpu+power supply is present.
3. Extra observations - my GPU in "idle" constantly blips the "throttling" flag on (for TEMP_HOTSPOT), even though the thing is at 40°C.
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
clean chroot building not flexible enough ?
Try clean chroot manager by graysky
Offline