Windows Research
On this page
Here we verify specific claims about Windows. In each article we state the verification conditions and the limitations of the result, and separately mark information from documentation, our own observations, measurements and their interpretation.
Before reading individual materials, review the page «How we research Windows». It explains the evidence statuses, measurement rules, data handling and the procedure for correcting conclusions.
Published research
Section titled “ Published research”New publications on top. Each study lists its publication date.
- Desktop vs sign-in screen: what a user session costs
5 hours of logged-in Windows 11 idle vs the sign-in screen: the first hour eats 79% CPU, the stationary desktop is only 1.6x noisier, and the Start feed keeps persistent connections.
- ETW autologgers in Windows 11: what actually writes to disk
39 autologger sessions in the registry, 19 active, but only 9 write to disk; the Diagtrack listener generates 110 MB of events in 2 hours, while 28 MB of the 32 are empty preallocated files.
- Placebo test: do subtle Registry tweaks actually suppress background activity
Strict A/B of 17 known "background reduction" settings on Windows 11: no combined effect, only LLMNR and Telemetry proven individually, three knobs disproven.
- How much memory can actually be freed in Windows
Measured limits of RAM freeing in Windows 11: disabling background components, terminating shell processes, cleaning the working set; standby is not a loss.
- Quiet idle: Windows background before and after optimization
Two independent measurement series on Windows 11: the effect of disabling background components on idle, the first minutes after boot, memory, and full CPU load.
- How to disable Cross-Device Resume
Cross-Device Resume research: why a Registry entry alone is not enough to disable it, what the MDM policy showed, and how to apply the setting in BoosterX.
- AFD, IRP stack and NetBIOS in Windows 11 25H2
Research on DisableRawSecurity, IrpStackSize and NetbiosOptions: system readers, ranges, scope and practical value.
- BoosterX security settings: what changes in Windows 11 25H2
Review of BoosterX security settings on Windows 11 25H2: actual policies, hardening and weakening of protection, limitations and recovery.
- DPC and Kernel Executive workers in Windows 11 25H2
Canonical DPC, worker thread, watchdog and scheduler assist parameters: readers, defaults, clamps and real-world usefulness.
- GPU presentation and power management in Windows 11 25H2
A study of ForceDirectFlip, GPU scheduler quantum and runtime power states: canonical paths, readers and hardware constraints.
- GPU TDR in Windows 11: TdrLevel, timeouts and retry limit
How dxgkrnl handles Windows 11 25H2 TDR parameters: canonical paths, defaults, clamps and the cost of changing GPU recovery.
- Keyboard port parameters Windows 11: MaximumPortsServiced and SendOutputToAllPorts
Research into kbdclass parameters for legacy keyboard ports: canonical path, defaults, reader, and scope of applicability.
- Lazy Writer, NTFS and Registry flush in Windows 11 25H2
How Windows 11 manages deferred writes, NTFS memory usage and the Registry flush interval: paths, defaults, readers and limitations.
- MMCSS, DWM and the Games profile in Windows 11 25H2
How Windows reads Multimedia Class Scheduler Service profiles for DWM and Games: values, defaults, timing and practical usefulness.
- Scheduler, timers and foreground boost in Windows 11 25H2
A study of Win32PrioritySeparation, timer policy, per-CPU ticks, DPC budgets and foreground boost: values, clamps and system readers.
- Service Host and Windows 11 background components
A study of svchost separation and background system components: Registry path, default, timing, resource impact, and limitations.
- Storage Sense, Start Search, and driver delivery in Windows 11
Documented Windows 11 policies for Storage Sense, web search, and automatic driver delivery: paths, defaults, and scope.
- TCP ACK and delayed ACK in Windows 11 25H2
TcpAckFrequency and TcpDelAckTicks for Windows interfaces: path, default, timing, and limits of impact on network applications.
- TCP connection lifecycle in Windows 11 25H2
Tcp1323Opts, TcpTimedWaitDelay, KeepAliveTime and PMTU black-hole detection: canonical paths, defaults and scopes.
- USB selective suspend and NVMe power states in Windows 11 25H2
How Windows and stornvme manage idle power for USB and NVMe: power scheme settings, defaults, GUIDs, and hardware limitations.
- Windows 11 privacy and security policies: Activity History, Defender, and WER
Canonical Windows 11 Registry policies for Activity History, Defender cloud lookup, Remote Assistance, Advertising ID, background apps, and WER.
- Windows Memory Manager and system cache: paging, page combining, and lazy write
Research into Memory Manager and Cache Manager Registry settings in Windows 11 25H2: canonical paths, defaults, clamps, readers, and practical usefulness.
- Windows shell and features: BoosterX settings on Windows 11 25H2
Which BoosterX settings change the Windows shell, notifications and optional features, what is outdated and what remains situational.
- Automatic Windows updates: drivers, maps, and Store apps
The boundaries of three Windows auto-update mechanisms: what gets disabled for drivers, maps, and Microsoft Store apps, and what keeps working.
- Windows 11 toast notifications: what gets disabled
Research into Windows 11 toast notification policies: app banners, lock screen, system messages, notification center, and outcome boundaries.
- SystemResponsiveness and MMCSS: what the values 0, 10, 20 and 100 do
A study of SystemResponsiveness and disabling MMCSS in Windows 11 25H2: value normalization, thread priority and synthetic scheduler latency.
- Background Raw Input listeners in Windows 10 and Windows 11
Why background Raw Input listeners in Windows 11 are capped at about 125 Hz and what physical and virtual WM_INPUT measurements showed.
- FSO, FSE and frame pacing in CS2
BoosterX research on how Fullscreen Optimizations and fullscreen exclusive affect Raw Input, click-to-photon latency and frame pacing in CS2.
- MPO in Windows: How Multiplane Overlay Works and When to Disable It
A study of Multiplane Overlay in Windows, OverlayTestMode=5, known driver issues, and safe diagnostics via GameReadyX.
- Scheduled Windows Diagnostics: what EnabledExecution disables
Static study of Scheduled Diagnostics: the EnabledExecution policy action, the limits of disabling it, and the absence of proven performance gains.
- Win32PrioritySeparation: latency and FPS at full CPU load
A study of foreground boost and Windows scheduler quanta with historical x300 click-to-photon measurements and FPS in Valorant.
- Windows Audio format: sample rate, bit depth and load
Measuring sample rate, bit depth, system effects, audio engine period and XAudio2 load on a single Realtek USB Audio endpoint.
How to read the conclusions
Section titled “ How to read the conclusions”- Documented means the behavior is described in a primary public source.
- Observed means the event was detected in the specified environment.
- Measured means a numerical value was obtained using the described methodology.
- Reproduced is used only after an independent repetition of the result.
- Insufficient data means the available observations do not allow a conclusion.
To extend a result to other Windows builds, computers or drivers, it must be repeated under those conditions.
Independence
Section titled “ Independence”The research is published by the BoosterX team and may relate to product features. The presence of a setting in BoosterX is not in itself considered proof of its effectiveness, and negative and inconclusive results are not excluded from the catalog.
BoosterX Wiki is an independent publication and is not affiliated with, authorized, sponsored or approved by Microsoft Corporation.
Pre-publication review
Section titled “ Pre-publication review”Before publication we check the source materials and public sources, confirm that changed settings are reverted, and remove personal and confidential data.