AMD P-State EPP Scaling Driver on AMD Ryzen 7 5800H laptop

[Nyakov13]

Ok, I think I got important finding’s, even crucial for me, and not sure where to put it.
Also, if someone have an idea where else I can address it, please let me know.

Probably this related to abrupt random shutdowns (power off) without any load, and without meaningful logs anywhere.

Kernel: 6.5.3-1-MANJARO
Hardware:
Model: HP Victus Laptop 16-e0xxx
CPU: AMD Ryzen 7 5800H
GPU: Radeon Vega Mobile and NVIDIA GeForce RTX 3060 6GB

The finding’s:

1. Kernel automatically loads AMD P-State EPP scaling (more info) driver (amd_pstate_epp).
2. By default, driver use “powersave” cpufreq governor (it is right thing to do, it allows performances hints from the OS, basically “performance” governor is useless for most of the cases)
3. By default, driver use “performance” hint (energy_performance_preference). And this leads to extremely high voltage on CPU(integrated GPU?) (vddgfx) in all times, as well as power consumption. 1.46V voltage and 16W power in totally idle state.
4. Setting energy_performance_preference to “balance_performance” leads to immediate drop of voltage to 1V and power draw to 5W
5. Setting energy_performance_preference to “power” drops voltage to 0.8V level and power draw to 3-4W.

My assumptions is, that first of all, it is extremely unhealthy to put (presumably integrated GPU?) to constant high voltage. And second, that under emerging load this causes immediate high power draw and quick chip heating without cooling system even active, and probably thermal shutdown of the system.

sensors output for default “performance” mode:

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:        1.46 V  
vddnb:       949.00 mV 
edge:         +43.0°C  
PPT:          15.00 W

sensors output for “balance_performance” mode:

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:      999.00 mV 
vddnb:       949.00 mV 
edge:         +43.0°C  
PPT:           4.00 W

Important!
power-profiles-daemon at the time of writing cannot use AMD P-State EPP amd_pstate_epp and platform_profile drivers at the same time, and, because platform_profile is available in system it will ignore amd_pstate_epp completely, leaving it in default values (“powersave” mode and “perfomance” energy preference)

To see in what mode now power-profiles-daemon, use powerprofilesctl command.
Related power-profiles-daemon issue.

Also, there is auto-epp python script/systemd service for automatically manage energy performance preferences depending on power source of laptop (AC or Battery)

You can set energy_performance_preference by doing:

# echo "balance_performance" | tee /sys/devices/system/cpu/cpufreq/policy*/energy_performance_preference

You can query available profiles by doing:

cat /sys/devices/system/cpu/cpufreq/policy0/energy_performance_available_preferences

To see current cpufreq state and temps:

# sensors
# cpupower frequency-info

Addition:
Ok. I installed zenpower3, and this is what I found.
On default “performance” preference I got constant 1.46V on CPU and integrated GPU.
This is BAD.

$ sensors
hp-isa-0000
Adapter: ISA adapter
fan1:           0 RPM
fan2:           0 RPM

nvme-pci-0500
Adapter: PCI adapter
Composite:    +37.9°C  (low  = -273.1°C, high = +80.8°C)
                       (crit = +81.8°C)
Sensor 1:     +37.9°C  (low  = -273.1°C, high = +65261.8°C)
Sensor 2:     +34.9°C  (low  = -273.1°C, high = +65261.8°C)

BAT0-acpi-0
Adapter: ACPI interface
in0:          16.81 V  

zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core:     1.46 V  
SVI2_SoC:    950.00 mV 
Tdie:         +41.6°C  (high = +95.0°C)
Tctl:         +41.6°C  
SVI2_P_Core:   9.64 W  
SVI2_P_SoC:    3.63 W  
SVI2_C_Core:   7.25 A  
SVI2_C_SoC:    3.83 A  

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:        1.46 V  
vddnb:       949.00 mV 
edge:         +41.0°C  
PPT:           8.00 W  

acpitz-acpi-0
Adapter: ACPI interface
temp1:        +41.0°C  (crit = +255.0°C)

On “balance_performance” I got dynamic voltage scaling and around 1V on CPU and GPU in idle state.

$ sensors
hp-isa-0000
Adapter: ISA adapter
fan1:           0 RPM
fan2:           0 RPM

nvme-pci-0500
Adapter: PCI adapter
Composite:    +36.9°C  (low  = -273.1°C, high = +80.8°C)
                       (crit = +81.8°C)
Sensor 1:     +36.9°C  (low  = -273.1°C, high = +65261.8°C)
Sensor 2:     +34.9°C  (low  = -273.1°C, high = +65261.8°C)

BAT0-acpi-0
Adapter: ACPI interface
in0:          16.81 V  

zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core:   1000.00 mV 
SVI2_SoC:    950.00 mV 
Tdie:         +43.1°C  (high = +95.0°C)
Tctl:         +43.1°C  
SVI2_P_Core:   3.27 W  
SVI2_P_SoC:    1.96 W  
SVI2_C_Core:   3.29 A  
SVI2_C_SoC:    2.06 A  

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:      999.00 mV 
vddnb:       949.00 mV 
edge:         +40.0°C  
PPT:           4.00 W  

acpitz-acpi-0
Adapter: ACPI interface
temp1:        +43.0°C  (crit = +255.0°C)

Useful links:

1. Benchmarks for Ryzen mobile system using AMD P-State EPP
2. Kernel.org documentation on new AMD P-State driver
3. power-profiles-daemon Consider using multiple “drivers” for CPU P-State support
4. CPU frequency scaling AchWiki
5. auto-epp python script
6. AMD P-State and AMD P-State EPP Scaling Driver Configuration Guide
7. Zenpower3

[cscs]

You might be interested in this thread:

Overheat or CPU Frequency Management Issues

I can offer a quick rundown of what I’ve done with a zen3 laptop: TLP ( not power-profiles-daemon ) zenpower ( zenpower3-dkms ) amd-pstate-epp ( using boot option amd_pstate=active ) Thats about it really. I believe yours is a zen2, but I believe all of the above is still compatible. I will also point out the pstate scenario is a bit confusing … and would take a lot of space here … but this reddit post seems roughly accurate: https://www.reddit.com/r/linux/comments/15p4bfs/amd_pstate_and_…

[Nyakov13]

Thank you for the info.
Although I do not want use TLP fro now because it is not the default and I am using KDE that have settings for power-profiles-daemon.
Need to look to zenpower3 probably… Not really want to install dkms…

[stephane]

so you can test

• amd-pstate=passive
• amd-pstate=active
• amd-pstate=guided

check with :
cpupower frequency-info and
sudo turbostats

in my case ( desktop + 5600x ) i use
“iommu=pt amd-pstate=passive nowatchdog processor.max_cstate=5 systemd.unified_cgroup_hierarchy=true scsi_mod.use_blk_mq=1”

which drivers videos are you using ?

[Nyakov13]

Ok. I installed zenpower3, and this is what I found.
On default “performance” preference I got constant 1.46V on CPU and integrated GPU.
This is BAD.

$ sensors
hp-isa-0000
Adapter: ISA adapter
fan1:           0 RPM
fan2:           0 RPM

nvme-pci-0500
Adapter: PCI adapter
Composite:    +37.9°C  (low  = -273.1°C, high = +80.8°C)
                       (crit = +81.8°C)
Sensor 1:     +37.9°C  (low  = -273.1°C, high = +65261.8°C)
Sensor 2:     +34.9°C  (low  = -273.1°C, high = +65261.8°C)

BAT0-acpi-0
Adapter: ACPI interface
in0:          16.81 V  

zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core:     1.46 V  
SVI2_SoC:    950.00 mV 
Tdie:         +41.6°C  (high = +95.0°C)
Tctl:         +41.6°C  
SVI2_P_Core:   9.64 W  
SVI2_P_SoC:    3.63 W  
SVI2_C_Core:   7.25 A  
SVI2_C_SoC:    3.83 A  

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:        1.46 V  
vddnb:       949.00 mV 
edge:         +41.0°C  
PPT:           8.00 W  

acpitz-acpi-0
Adapter: ACPI interface
temp1:        +41.0°C  (crit = +255.0°C)

On “balance_performance” I got dynamic voltage scaling and around 1V on CPU and GPU in idle state.

$ sensors
hp-isa-0000
Adapter: ISA adapter
fan1:           0 RPM
fan2:           0 RPM

nvme-pci-0500
Adapter: PCI adapter
Composite:    +36.9°C  (low  = -273.1°C, high = +80.8°C)
                       (crit = +81.8°C)
Sensor 1:     +36.9°C  (low  = -273.1°C, high = +65261.8°C)
Sensor 2:     +34.9°C  (low  = -273.1°C, high = +65261.8°C)

BAT0-acpi-0
Adapter: ACPI interface
in0:          16.81 V  

zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core:   1000.00 mV 
SVI2_SoC:    950.00 mV 
Tdie:         +43.1°C  (high = +95.0°C)
Tctl:         +43.1°C  
SVI2_P_Core:   3.27 W  
SVI2_P_SoC:    1.96 W  
SVI2_C_Core:   3.29 A  
SVI2_C_SoC:    2.06 A  

amdgpu-pci-0600
Adapter: PCI adapter
vddgfx:      999.00 mV 
vddnb:       949.00 mV 
edge:         +40.0°C  
PPT:           4.00 W  

acpitz-acpi-0
Adapter: ACPI interface
temp1:        +43.0°C  (crit = +255.0°C)

Also. No random shutdowns till this moment!

[ling]

After reading these posts, I’m a bit confused. Is your solution simply to adjust the system’s power mode to balanced?

I can describe in detail the issue I encountered:
Initially, when I first installed Manjaro, my computer would freeze under low load and required a long press of the power button to restart (similar to your situation). Later, I tried limiting cstate to 5 (because someone mentioned that AMD c6 could cause this issue), as well as trying other Linux boot parameters, but none of them worked. My final solution was to open a virtual machine in VirtualBox to prevent the computer from entering a low-power state.

This solution has been effective all along, and this month I noticed that even if I don’t open the virtual machine, the freezing issue has disappeared. This is quite surprising because I didn’t spend any more time on this issue. Based on your post, I believe that after a system update, my power management was able to adjust to balanced mode normally, as I always set it to default to balanced upon startup.

Until last week, when I upgraded to Linux kernel 66 (because 65 is no longer maintained), I found that the computer would randomly restart without any warning. Please note that it’s not freezing, but an automatic restart.

After this issue occurred, I tried to find a solution but found no useful information. Therefore, I had to downgrade the Linux kernel to 61 (I also tried booting into 65, but it still caused restarts, possibly because some drivers were also upgraded during the 65 upgrade).

Kernel 61 worked fine without any problems.

Just now, following your post, I set the power mode to power-saver, and shortly afterward, the issue reappeared: automatic restarts.

So, I’m glad to have found the cause of the random restarts last week, and I appreciate your work here!

However, I hope to find a perfect solution: the computer should work normally in any mode. What should I do? Thanks everyone for your answers!

[cscs]

@ling

Is your solution simply to adjust the system’s power mode to balanced?

We were discussing the AMD P-State driver … which can control your available ‘power modes’ and when they are switched.

This can work in conjunction with power management software like TLP or power-profiles-daemon.

As far as I can tell from the thread … OP did not mark any solution.

@ling

Just now, following your post, I set the power mode to power-saver, and shortly afterward, the issue reappeared: automatic restarts.

That isnt suggested here.

@ling

What should I do?

Sounds like you should get a handle on your systems configuration.

It also sounds like maybe you should not be using power-profiles-daemon and/or whatever scaling driver you are using.

This is all highly dependent on your actual hardware (and use case).

[ling]

@cscs

That isnt suggested here.

Oh, Perhaps my expression was misunderstood. What I meant was that doing so caused the issue to reoccur for me. However, afterward, I still set the power management to balanced to ensure the system runs normally.

@cscs

This is all highly dependent on your actual hardware (and use case).

Thank you for your suggestion.

[Tanker1]

I found a simple solution to keep stability of manjaro to avoid the impact of c6 state on my computer. The similar problem didn’t occur after I’m running command ‘/usr/bin/stress --cpu 1’ to prevent cpu usage from lower level.

[cscs]

Wait, so you ran stress on ‘cpu 1’ and then everything is fine?
Or you are constantly stressing ‘cpu 1’?

Is this really preferable to the outlined c6 workarounds?
https://wiki.archlinux.org/title/Ryzen#Soft_lock_freezing

And is this really the same issue as OP here?

[Tanker1]

@cscs

constantly stressing

Constantly stressing cpu. And if i stopped stressing, it freezed again. I’ve disabled c6 state and processor.max_cstate=1 according to the link you posted but it didn’t work. By the way, my UEFI didn’t have an option “Power idle control”.

[cscs]

@Tanker1

I’ve disabled c6 state and processor.max_cstate=1 according to the link you posted but it didn’t work

Link does not suggest that.

This bug is well known and is being discussed on bugzilla and launchpad. While the solution is not the same in all cases, this one helped some users. Add the output of this command echo rcu_nocbs=0-$(($(nproc)-1)) as a kernel parameter where the command nproc just prints your CPU’s threads. For this option to be applied, you need a compiled kernel with option CONFIG_RCU_NOCB_CPU (like linux).

A different cause for the freezes is the power saving management indicated by c-states. The maximum power saving state c6 can cause problems. Adding the kernel parameter processor.max_cstate=5 helped in some cases

And was it properly applied? Like you updated grub afterwards?

The link also offers secondary advice in the case it somehow does not work.

but other users reported that the option is not applied and the c6 state is still entered. For them, this package disable-c6-systemdAUR helped. Before using it, modprobe msr needs to be run in order to activate that kernel module.

Some laptops with Ryzen CPUs such as the HP Envy x360 15-bq100na may experience CPU soft locks which result in a frozen system. These can be avoided with the kernel parameter idle=nomwait added.

In some cases, kernel parameter pci=nomsi fixes the issue.

Finally in the worst case scenarios…

In some other cases, the issue is simply bad hardware, and warranty claiming the CPU for a new one may just solve your issues.

But also note none of that is explicitly linked to the scaling driver.

I still fail to see how your system freezing (possibly by c state) is related to this thread.

[Tanker1]

@cscs

Like you updated grub afterwards

Sorry for jumbling of this original situation of this thread and my circumstance. Here is the grub configuration ‘GRUB_CMDLINE_LINUX_DEFAULT=“quiet splash udev.log_priority=3 idle=nomwait processor.max_cstate=1 intel_idle.max_cstate=0”’ I tried. It differs from yours so I’ll applied yours properly next time. My disable-c6 configuration is consistent with the link.
Unfortunately, the solutions of the link didn’t work absolutely unless substituting the amd with intel or stressing the cpu constantly, after I implemented what you said.

[ling]

I have a better method to prevent the CPU from entering low-power mode: Install VirtualBox and run a virtual machine in the background, such as Windows 11, since there are still some applications that require Windows. This way, the CPU will remain active without wasting resources like the ‘stress’ utility would. I have done this, and it’s useful.

[Tanker1]

Thanks for your advice. I’m trying.