I have an M.2 A+E key Coral module installed in a Lenovo M920q PC running TrueNAS, with the latest BIOS and all tuning suggestions from powertop applied.
Without the TPU installed, the PC can spend most of its time in package C-state PC8, but with the TPU present and fully idle, the PC doesn't go below C-state PC3. Simply having the TPU installed adds about 3 watts to the PC's idle power consumption. This is significantly above the datasheet 375-400mW idle power for the TPU itself, and I suspect can mostly be attributed to the PC being unable to enter lower C-states. lspci -vvv reports that the TPU supports ASPM L0s/L1 but that it's disabled, and the same appears with pcie_aspm=force on the kernel command line.
I've seen a few references to manually disabling ASPM in other bug reports, but I'm not sure if that's as a workaround for ASPM bugs in host systems or due to issues with the Coral device itself. So... is this expected to work and the issue is some kernel or BIOS misconfiguration on my side, or does the Coral just not support ASPM?
Click to expand!
### Issue Type
Support
### Operating System
Linux
### Coral Device
M.2 Accelerator A+E
### Other Devices
_No response_
### Programming Language
Other
### Relevant Log Output
_No response_
Description
I have an M.2 A+E key Coral module installed in a Lenovo M920q PC running TrueNAS, with the latest BIOS and all tuning suggestions from
powertopapplied.Without the TPU installed, the PC can spend most of its time in package C-state PC8, but with the TPU present and fully idle, the PC doesn't go below C-state PC3. Simply having the TPU installed adds about 3 watts to the PC's idle power consumption. This is significantly above the datasheet 375-400mW idle power for the TPU itself, and I suspect can mostly be attributed to the PC being unable to enter lower C-states.
lspci -vvvreports that the TPU supports ASPM L0s/L1 but that it's disabled, and the same appears withpcie_aspm=forceon the kernel command line.I've seen a few references to manually disabling ASPM in other bug reports, but I'm not sure if that's as a workaround for ASPM bugs in host systems or due to issues with the Coral device itself. So... is this expected to work and the issue is some kernel or BIOS misconfiguration on my side, or does the Coral just not support ASPM?
Click to expand!
### Issue Type Support ### Operating System Linux ### Coral Device M.2 Accelerator A+E ### Other Devices _No response_ ### Programming Language Other ### Relevant Log Output _No response_