XPS 8960, making continuous clicking noises when in sleep mode
As soon as the XPS 8960 enters sleep mode, the PSU starts making clicking noises. The clicking continues indefinitely, at a rate of one click every 2 seconds, until the computer is awaken.
The PSU's green light at the back of the box turns on and off in sync with the clicking sound. The clicking itself sounds like a mechanical relay inside the PSU opening and closing continuously, which can't be good.
Here's a video showing this behavior. The computer goes to sleep at timestamp 1s, and is woken up at timestamp 25s:
Alright, after a bit of experimentation, I finally figured out what's going on, as well as two workarounds.
Under both Windows and Linux, the box draws around 45W when it's powered on but otherwise inactive.
When suspended under Windows, the power drops to about 25W after 30 seconds, then down to ~12W after 90 seconds and stabilizes at that level until resumed, except for some brief random spikes anywhere between 15W and 100W now and again.
In contrast , when suspended under Linux, the power drops almost immediately to a steady 6-8 watts, and stays there until resumed. So Linux clearly goes into a deeper sleep.
According to its ACPI tables, in the S0 sleep state (aka Modern Standby) the Dell New XPS 8960 supports four idle power states: POLL, C1_ACPI, C2_ACPI, C3_ACPI. Using the intel_idle.max_cstate kernel boot parameter to restrict the idle power states allowed for the CPUs, we can measure how the power usage varies with the idle power state during suspend:
Max_cstate
Power
POLL
45W
C1_ACPI
24W
C2_ACPI
12W
C3_ACPI
6-8W
This suggests that when suspending, Windows first goes to C1 after 30 seconds, then to C2 after 90 seconds and stays there until resumed, while Linux goes directly to C3 (which is a valid state according to the ACPI tables) and stays there until resumed.
In that experiment, the continuous clicking problem only shows up when suspending in C3, which is why it doesn't reproduce on Windows, because Windows only ever goes up to C2.
To confirm that this is a C3 issue, and not a power threshold issue (e.g. when the power draw drops below ~10W), we enable PowerShare and charge a device via a PowerShare enabled USB ports, to increase the load comfortably above 12 watts. The clicking continues when suspended in C3, with a >12W load proving that the cause is C3, not a power draw on the PSU that's too low.
Workaround #1:
Given the observations above, we work around the issue by simply disabling C3 before suspend, effectively forcing sleep at C2, and re-enabling it after resume:
# cat > /usr/lib/systemd/system-sleep/xps-8960-psu-fix.sh <<-"EOF"
case "$1" in
pre)
echo "Disabling C3_APCI (workaroud for PSU clicking while in suspended state)"
cpupower idle-set --disable 3
;;
post)
echo "Reenabling all Cx_ACPI states"
cpupower idle-set --enable-all
;;
esac
EOF
# chmod a+x /usr/lib/systemd/system-sleep/xps-8960-psu-fix.sh
Workaround #2:
Digging a bit deeper, we find that the issue comes from these two Dynamic OEM ACPI tables:
Whenever they're loaded, the clicking happens, whenever they aren't, it doesn't. We can prevent them from loading via this kernel boot option:
acpi_osi=!
As a bonus, with this method, the power usage during suspend drops a bit below 10W.
These tables contain only a modest amount of ACPI firmware code. However I'm not entirely sure about all that this code does. So workaround #1 might be a safer approach. In any case, with either of these workarounds, the clicking problem vanishes.
Conclusion:
Please note that as this is a Linux-based issue, it is outside of our scope as it is not shipped with our system. We will not be able to provide further assistance or support for this issue
This investigation proves that the issue did not come from Linux, but from buggy ACPI tables which are very much shipped with the system and therefore Dell's responsibility. Something that can of course only be proven by actually investigating, yet was refused by Dell Support.
Which begs the question: is the #1 PC vendor really in the business of supporting only the #1 desktop OS on its hardware?
I contacted Dell Tech Support. After reading my description, and watching the video, they mentioned having seen similar PSU problems with other customers, and recommended replacing the PSU.
They kept mentioning the fan though, which seemed odd, given the symptoms. The main clicking sound sounds a lot more like a mechanical relay opening and closing. The PSU's green light going on and off in sync with that loud sharp clicking kind of confirms that. I expressed my doubts wrt the fan related diagnosis, but they were confident that replacing the PSU would address the issue.
Anyways, a Dell technician replaced the PSU today. The problem persists identically with the brand new power supply. I demonstrated it to the technician. So I'm going back to Dell Tech Support for the next steps.
This may have something to do with the storage controllers. The symptoms change with the "SATA/NVMe Operation Mode" BIOS setting:
When setting the "SATA/NVMe Operation Mode" to "AHCI/NVMe", the system misbehaves as described earlier in this post. Interestingly, whether there is any drive actually plugged in or not
When setting the "SATA/NVMe Operation Mode" to "RAID on", the systems suspends and wakes up normally, but only if left suspended for less than 45 seconds. When suspended for 60 seconds or more, the system either never wakes up (and has to be hard powered off) or reboots instead of waking up gracefully. Between 45 and 60 seconds, it's a roll of the dice: the systems wakes up more often than not, but not always. Not that it really matters, as this sleep duration is useless for all practical purposes.
"Please note that as this is a Linux-based issue, it is outside of our scope as it is not shipped with our system. We will not be able to provide further assistance or support for this issue and we would like to inform you that we are cancelling the mail in service which we have created.
If you have any further questions or concerns, please don't hesitate to reach out. We appreciate your understanding and look forward to assisting you with any other matters.
For XPS 8950/8960; Aurora R13/R14/R16; replacement 8YJCT Liteon 1000W PSU, R61D8 Liteon 1000W PSU, RD0G0 Delta 1000W PSU will not change this. This is an expected behavior with these high-efficiency PSU. The clicking noise is a relay operation in the high-power and high-efficiency PSU. The relay is part of the power-saving design when turning on or turning off the PSU. The relay turns on or off to produce a metal clicking sound.
My complaint is not about the PSU having a relay that clicks once when the box powers on or off. It's that it does so continuously every two seconds (as shown in the video) for as long as the computer remains in suspended mode. I have a hard time believing that this is by design. If it is however, and is deemed up to and representative of Dell's design standards, then I have questions about these standards.
Checking residency of the various idle power states shows that workaround #2 actually uses C3, which explains why its power usage is better than workaround #1, and further confirms that the issue is with whatever these two buggy ACPI tables do when in C3, not with the handling of C3 by the OS.
fozzy8900
1 Rookie
•
18 Posts
0
2
Posted June 29th, 2024 04:39
Alright, after a bit of experimentation, I finally figured out what's going on, as well as two workarounds.
Under both Windows and Linux, the box draws around 45W when it's powered on but otherwise inactive.
When suspended under Windows, the power drops to about 25W after 30 seconds, then down to ~12W after 90 seconds and stabilizes at that level until resumed, except for some brief random spikes anywhere between 15W and 100W now and again.
In contrast , when suspended under Linux, the power drops almost immediately to a steady 6-8 watts, and stays there until resumed. So Linux clearly goes into a deeper sleep.
According to its ACPI tables, in the S0 sleep state (aka Modern Standby) the Dell New XPS 8960 supports four idle power states: POLL, C1_ACPI, C2_ACPI, C3_ACPI. Using the intel_idle.max_cstate kernel boot parameter to restrict the idle power states allowed for the CPUs, we can measure how the power usage varies with the idle power state during suspend:
This suggests that when suspending, Windows first goes to C1 after 30 seconds, then to C2 after 90 seconds and stays there until resumed, while Linux goes directly to C3 (which is a valid state according to the ACPI tables) and stays there until resumed.
In that experiment, the continuous clicking problem only shows up when suspending in C3, which is why it doesn't reproduce on Windows, because Windows only ever goes up to C2.
To confirm that this is a C3 issue, and not a power threshold issue (e.g. when the power draw drops below ~10W), we enable PowerShare and charge a device via a PowerShare enabled USB ports, to increase the load comfortably above 12 watts. The clicking continues when suspended in C3, with a >12W load proving that the cause is C3, not a power draw on the PSU that's too low.
Workaround #1:
Given the observations above, we work around the issue by simply disabling C3 before suspend, effectively forcing sleep at C2, and re-enabling it after resume:
Workaround #2:
Digging a bit deeper, we find that the issue comes from these two Dynamic OEM ACPI tables:
Whenever they're loaded, the clicking happens, whenever they aren't, it doesn't. We can prevent them from loading via this kernel boot option:
As a bonus, with this method, the power usage during suspend drops a bit below 10W.
These tables contain only a modest amount of ACPI firmware code. However I'm not entirely sure about all that this code does. So workaround #1 might be a safer approach. In any case, with either of these workarounds, the clicking problem vanishes.
Conclusion:
This investigation proves that the issue did not come from Linux, but from buggy ACPI tables which are very much shipped with the system and therefore Dell's responsibility. Something that can of course only be proven by actually investigating, yet was refused by Dell Support.
Which begs the question: is the #1 PC vendor really in the business of supporting only the #1 desktop OS on its hardware?
(edited)