Since this request was mentioned on T3, I wanted to follow up with results from my own lab now that I’ve got intel_idle working properly on TrueNAS SCALE.
** Setup**
- Motherboard: ASUS Z790
- CPU: Intel i7-13700K
- Drives: 15 × HDDs, 6 × SSDs, 3 × NVMe
- Fans: 7 case + 2 CPU + 2 HBA (40mm)
- GPU: RTX 4070 SUPER
- OS: TrueNAS SCALE 25.04.2
- PowerTOP: Compiled from source (v2.15)
- Driver:
intel_idle now active
- ASPM: L1 enabled on all relevant devices
- Note: A
zfs send was running during all power measurements
** Power Comparison**
Before (acpi_idle): ~145W idle
After (intel_idle): ~127W idle
This 18W drop was measured during a ZFS send.
Important context: These readings reflect a system currently running with 7 HDDs (1 temporarily removed), plus all 6 SSDs and 3 NVMe drives. Final drive count will increase.
Deepest idle state:
Before: C3_ACPI
After : PC10 (C10) with up to 85% residency per core
Package sleep state:
Before: Pkg(OS) 0.0%
After : Pkg(HW) active in C10
** Notes**
I compiled PowerTOP from source in a Bookworm LXC, copied it to /mnt/tank/tools/powertop , and created a systemd service for auto-tune on boot. No post-init task needed.
Why PowerTOP 2.15 Matters
TrueNAS SCALE ships with PowerTOP v2.14, which was released before September 2022.This outdated version:
❌ Cannot detect C10 residency properly
❌ Fail to auto-tune ASPM or PCIe L1 reliably
❌ Miss hybrid-core optimizations on 12th–14th Gen Intel CPUs
Meanwhile, PowerTOP v2.15 was released on September 29, 2022, and it’s the first version that supports: chough cough 3 years ago?
✅ Alder Lake and Raptor Lake CPUs
✅ Accurate ASPM detection on Z690/Z790 chipsets
✅ Reliable tuning on Linux 6.x kernels (like SCALE 25.04.2)
✅ Deep C-State visibility (C10, Pkg(HW), etc.)
Bottom line: PowerTOP 2.14 is too old to be useful for modern platforms. If you’re on TrueNAS SCALE with Intel 12th+ Gen, compiling v2.15 from source is mandatory for accurate tuning.
The key fix is that intel_idle is available in SCALE if your system supports it:
cat /sys/devices/system/cpu/cpuidle/current_driver
intel_idle
And ASPM is working across all bridges:
lspci -vv | grep -i ASPM
Special thanks to @rs443 and @Salaman for the help.
This change alone shaved ~18W off idle power, and I expect that to improve further once idle stabilizes post-transfer.
** Estimated Power Savings**
After consolidating drives and powering off the Proxmox box (83W draw):
Total saved: 83W (Proxmox box) + 18W (intel_idle gain) = 101W
Annual usage: 101W × 24h × 365 = 884.76 kWh
At $0.13/kWh: 884.76 × 0.13 = $115.02/year
** Annual savings: ~$115.02**
** Additional Power Draw Estimate**
Adding 6 more HDDs to reach the full 15 will likely increase idle power by 24W, or **$27.33/year** at $0.13/kWh.
However, that will be offset by eliminating the second server (Proxmox box) and its 83W draw. Once I finish consolidating everything into my HL15, I’ll update the numbers with live readings.
