Starting with Linux kernel 6.7, users of the AMDGPU driver are not be able to set power limits below the recommended values advised by the AMD Engineering team on the hardware itself. The new low-power limits are intentionally enforced and set based on each card vBIOS specification.
While before the recent 6.7 kernel release the maximum power limit was on-par with what is specified at the vBIOS, but the driver didn't enforce any lower power limits, making the driver really good for fine-tuning it for power saving purposes. However, setting lower power limits that didn't match with what is specified at the vBIOS might harm the GPU as stated on the bug report that lead to this implementation:
This has been discussed on the amd-gfx M/L, and the conclusion was that under-powering outside of the bounding box is potentially dangerous and might damage the hardware. This won't be added back in.
And from the amd-gfx mail list:
The change aligns the driver what has been validated on each board design. Windows uses the same limits. Using values lower than the validated range can lead to undefined behavior and could potentially damage your hardware.
The other interesting information extracted from that mail list discussion is that the AMDGPU Linux driver will basically work the same way as the Windows driver, obeying the values stated at vBIOS level.
Opinion: While the changes seem frustrating at first glance for those who want some extra power saving features for computers running on idle, this is an open source driver and it will likely not take long for third party repositories (aur, ppa, copr) to distribute unpatched versions of this module. This already happened in the past with ZFS, NVIDIA, etc. Although, as mentioned, you run the risk of hardware damage if you do decide to use any that pop up.
Article taken from GamingOnLinux.com.
I don't like this kind of overbearing handholding. It's fine to have the driver adhere to the board manufacturers specifications, it's another thing to block functionality and outright refuse to entertain the option. That's not cool.
Because people are upset over this regression and feature limitation, a patch has been created to restore it by user fililip in Gitlab issue 3183
Quoting: dpanterThis is a bit of unfortunate nonsense, if users want to break their hardware then they should be able to. Not that lowering power limit actually does that but whatever.
I don't like this kind of overbearing handholding. It's fine to have the driver adhere to the board manufacturers specifications, it's another thing to block functionality and outright refuse to entertain the option. That's not cool.
Because people are upset over this regression and feature limitation, a patch has been created to restore it by user fililip in Gitlab issue 3183
While I agree with your point that customers ultimately should OWN the hardware in all levels, most of amdgpu developers are AMD Engineers, and they seem to be defending the company against any hardware damage that might lead to warranty issues.
I'm not saying that I agree with them, but I kinda understand the reason. Being a Support Engineer myself, I had to face a couple of those situations where the Engineering team implements some lock mechanisms and I had to deal with the splash damage of explaining the customer some "brand new limitation on our shiny software".
Beta Version Mar 5
Linux_Rocks Mar 5
Quoting: chtkSo, wait. The current stable release is 6.7 (patch 8). Will this be patched in 6.7 or the currently in rc 6.8?
It is already applied. While I don't have an AMD GPU to test, there are these 2 reports from 1 month ago in this issue that reinforce this information:
This comment:
Since kernel 6.7, power1_cap_min went from 0 to 136000000 on my 6600XT.
I understand this is the behavior introduced in commit 1958946858a62b6b5392ed075aa219d199bcae39.
Am I understanding correctly that this is a safety feature to prevent the over-draw issue?
I used to set my power1_cap at 75W (as a hard limit on noise and heat) and it worked fine for me.
That's now not possible due to the higher power1_cap_min.
Was this always out of spec and I was just lucky to not encounter the issue?And this comment:
Dear maintainers, I too am affected by the issue described above, i.e. power1_cap_min went to 96W on my RX6600
(default is 100W I think); before, I could happily apply a 50W power cap. Is there any possibility to restore
the ability to set lower power caps?Last edited by nwildner on 5 March 2024 at 5:34 pm UTC
michaldybczak Mar 5
Is it only about dGPU or maybe also iGPU?
Usually, dGPU is powered down (or shows energy usage below 1W). When it is activated for games, then I guess it should deliver whatever it has in store, but otherwise, it should power down again, when it's no longer used. However, I'm not sure how it works when low demanding titles are launched.
In my case, I use TUXEDO laptop and I use Tuxedo Control Center (in short TCC) and it has power profiles. What it does (from user perspective), is that it can limit the CPU frequency (not the power usage per se), so the heat production and energy usage becomes lower. My question is, will this limitation influence TTC or maybe it has nothing to do with it?
In other words, on a daily basis, I don't need as much power and like to use low demand power profiles (limits CPU 1,4GHz). System words as fast as usual, but it doesn't have CPU spikes, doesn't overheat and fans are not working so frequently. It is an awesome feature and I'm afraid it won't work as good now. Or maybe my fears are unwarranted?
136W minimum for a 6600XT surely sounds a bit too high for a Minimum. On nvidia at least usually the limit ranges are sane. Higher end GPUs have higher minimum thats allowed.
RTX 3080 for example:
GPU Power Readings
Power Draw : 37.30 W
Current Power Limit : 370.00 W
Requested Power Limit : 370.00 W
Default Power Limit : 370.00 W
Min Power Limit : 100.00 W
Max Power Limit : 370.00 WLast edited by Xpander on 6 March 2024 at 8:36 am UTC
Quoting: michaldybczakHow this is plays on hybrid setup AMD CPU/AMD GPU?
Is it only about dGPU or maybe also iGPU?
Usually, dGPU is powered down (or shows energy usage below 1W). When it is activated for games, then I guess it should deliver whatever it has in store, but otherwise, it should power down again, when it's no longer used. However, I'm not sure how it works when low demanding titles are launched.
I think that on Laptop usecases the impact will not be THAT big because usuallt the lower voltages are way lower on those dGPUs than they are on Desktop GPUs.
My Laptop RXT3070 can get as low as 8W on Idle when I'm on battery and set for optimus mode looking at nvidia-smi.
But since even Firefox tries to use GPU nowadays by setting GL_ instructions, I end up configuring my laptop into "iGPU only" on battery which will send a pci detach to the nvidia dGPU making it unavailable and completely powered off.
Last edited by nwildner on 6 March 2024 at 11:05 am UTC
Patreon
PayPal
See more from me