This Is Win 11

windowswhisperer

Windows 11 appears to dynamically override the power plan’s Processor Energy Performance Preference (EPP) in response to per-thread/process QoS hints (EcoQoS/Efficiency mode), Game Mode, and heterogeneous scheduling heuristics. Looking for authoritative guidance on auditing and controlling this behavior without disabling modern scheduling.

Context

OS: Windows 11 (23H2/24H2), all firmware mitigations enabled (VBS/HVCI), hybrid CPUs (Intel P/E-cores or AMD CPPC2).
Power plan: Balanced, with EPP set explicitly via powercfg to a known value (e.g., 0×32/50).
Features in play: Efficiency mode (Task Manager), EcoQoS APIs, Game Mode, Modern Standby (S0ix), Hardware Accelerated GPU Scheduling.

Observed behavior

The configured EPP value for the active scheme (GUID 36687f9e-e3a5-4dbf-b1dc-15eb381c6863) is honored at boot, but the effective per-core EPP fluctuates at runtime when:
- A foreground game is detected (Game Mode on).
- Background processes enter Efficiency mode or use EcoQoS.
- The system transitions in/out of Modern Standby or toggles Energy Saver.
Adjustments do not persist in the power plan UI and do not change the stored AC/DC EPP indexes, but they affect performance scaling immediately (confirmed via ETW counters and performance limit telemetry).
On hybrid CPUs, the OS also shifts load between P/E cores in a way that correlates with these EPP changes, complicating reproducible benchmarks.

Repro (one example)

Set Balanced EPP: powercfg /SETACVALUEINDEX SCHEME_BALANCED SUB_PROCESSOR PERFENERGYPREFERENCE 50; powercfg /S SCHEME_BALANCED.
Start an app and enable Efficiency mode in Task Manager, then observe:
- ETW providers Microsoft-Windows-Kernel-Processor-Power and Microsoft-Windows-Power indicate policy updates.
- Processor performance/p-state behavior changes despite no visible power plan setting change.
Toggling Game Mode produces a different effective EPP envelope and thread migration pattern even with the same plan.

Questions
1) Is there a supported mechanism to “lock” effective EPP for the active plan during testing (e.g., a policy switch or documented registry key) that prevents dynamic overrides from EcoQoS/Game Mode, without disabling heterogeneous scheduling or turning off Thread Director/CPPC2?
2) What is the authoritative telemetry path to attribute an effective EPP change to its source?

ETW providers/events that map a delta to: Game Mode, EcoQoS (SetThreadInformation/ProcessPowerThrottling), Energy Saver, Connected Standby, OEM/ACPI DPTF/PPM policies, or OEM control apps.
Any specific event IDs or fields that expose old/new EPP and the initiating component.
3) Does Game Mode on Windows 11 adjust only scheduling and GPU policy, or does it also modify HWP/CPPC EPP globally or per-core? If so, is the mapping documented?
4) Can EcoQoS be scoped so it never escalates/lowers global EPP-i.e., strictly a per-thread/per-process hint-via policy or API flags? Conversely, can it be disabled system-wide via Group Policy or MDM while retaining Efficiency mode UI?
5) Do Hyper-V, WSL2, or VBS materially influence EPP and processor performance states in Windows 11, or are any observed shifts purely OS policy above the virtualization layer?
6) For reproducible performance testing, what is the recommended Windows 11 configuration to:
Fix EPP and min/max processor performance,
Neutralize Game Mode and EcoQoS effects,
Maintain hybrid scheduling,
And verify via ETW or perf counters that the configuration is stable across sleep/resume?

What I’ve tried

Fixing EPP via powercfg and validating the stored value; effective behavior still shifts.
Disabling Game Mode; reduces but does not eliminate dynamic policy changes.
Monitoring ETW: Microsoft-Windows-Kernel-Processor-Power, Microsoft-Windows-PDC, Microsoft-Windows-Power, Microsoft-Windows-ThreadScheduling; can see policy transitions but lack source attribution clarity.
Confirming that OEM power services are stopped; behavior persists on clean installs.

Looking for documentation or field-proven procedures from those who have instrumented this on Windows 11, including exact ETW event names/IDs, relevant Group Policies/MDM CSPs (if any), and a minimal set of changes to achieve a stable, testable power/perf baseline without disabling hybrid-aware scheduling.

digimaven

A few things that helped me nail this down for repeatable runs on 23H2/24H2 hybrid boxes:

There isn’t a supported switch that “freezes” effective EPP while keeping heterogeneous scheduling + CPPC2/Thread Director. Windows 11’s Effective Power Mode (EPM) aggregator can still bias EPP in response to Game Mode, Energy Saver, and EcoQoS. The only reliable way I’ve found to stop the EPP drift is to turn off autonomous control and boost in the plan (keeps hybrid scheduling intact, but you’re trading off HWP/CPPC autonomy):
- In your active scheme, set “Processor performance autonomous mode” = Disabled and “Processor performance boost mode” = Disabled (powercfg /Q to grab the exact GUIDs on your build).
- Keep your EPP, min/max processor perf, and parking settings fixed after that. This has been the most stable baseline for A/B testing without killing P/E-core awareness.
Attribution trail (short list that’s actually useful in WPA):
- Use WPR “Power” + “CPU” profiles, then in WPA:
- Power > Power Settings table: add columns “Power Setting GUID”, “New Value”, “Attribution” (this one is key), “Process”, “Stack”. You’ll see changes tied to GUID 36687f9e-… for EPP and who initiated it.
- Power > Effective Power Mode timeline: shows shifts to Balanced/HighPerformance/GameMode. That’s the aggregation layer that later drives EPP nudges.
- Computation > CPU > CPU Performance (HWP/CPPC): add “EPP” and “Reason”/“Source” columns (on Intel this surfaces HWP request updates; on AMD you’ll see CPPC requests).
- Providers to include if you roll your own ETW: Microsoft-Windows-Power (PowerSettingChange, EffectivePowerModeChange), Microsoft-Windows-Kernel-Processor-Power (HWP/CPPC request updates, perf policy changes), Microsoft-Windows-ThreadScheduling (QoS/EcoQoS transitions), Microsoft.Gaming.GameBar (Game Mode toggles). The WPA “Attribution” field generally saves more time than memorizing event IDs.
Game Mode absolutely does more than GPU and thread priority on Win11. It elevates the system’s Effective Power Mode, which in turn biases EPP and perf ceilings for the foreground session. The exact EPP mapping isn’t publicly documented and varies by OEM policy/PEP.
EcoQoS scoping/disable:
- There’s no policy to force EcoQoS to be “strictly local” or to block it system-wide while keeping Efficiency mode UI. The “Turn off Power Throttling” GPO/MDM (HKLM\SYSTEM\CurrentControlSet\Control\Power\PowerThrottling\PowerThrottlingOff=1) stops OS-initiated background throttling but does not prevent apps or Task Manager from setting EcoQoS hints.
- If you need to neutralize it during tests, disable Power Throttling as above and avoid using Efficiency mode; for stricter control, test under a separate user context without Task Manager privileges or block taskmgr via WDAC/AppLocker for the test account.
Hyper-V/WSL2/VBS: they don’t directly change EPP policy. You’ll see minor noise from different timer behavior/interrupt distribution, but EPP/CPPC requests come from the host policy layer, not the hypervisor.
A compact “stable baseline” recipe I’ve been using:
- Power plan: High performance (or Balanced if you must), set EPP to your target; set min/max processor perf to desired fixed values.
- Disable “Processor performance autonomous mode” and “Processor performance boost mode.”
- Disable Game Mode (GPO: Windows Components > Game Bar > Allow Game Mode = Disabled or HKCU\Software\Microsoft\GameBar\AllowAutoGameMode=0).
- Disable Power Throttling (GPO or HKLM…\PowerThrottling\PowerThrottlingOff=1).
- Turn off Energy Saver/Battery Saver and set “System power mode” to Best performance.
- Leave hybrid scheduling on; don’t touch core parking if you want hetero behavior, but keep the parking min/max fixed.
- Verify: WPR Power+CPU, check WPA Power Settings + Effective Power Mode + CPU Performance (HWP/CPPC). Make sure EPP stays flat across S0ix cycles in the trace.

If you’ve got a WPA screenshot where EPP steps at the exact moment a process flips to EcoQoS, can you share the “Attribution” column value and the Effective Power Mode timeline around that point? That’s usually where the smoking gun shows up.