NOVA only ships documented, reversible changes, each with an evidence rating. The popular tweaks below are left out on purpose: measurements show they don't help, or they cost stability or security.
NOVAbcdedit (useplatformclock, disabledynamictick, useplatformtick, tscsyncpolicy)Why we don't: forcing HPET makes timer queries more expensive, community measurements are inconsistent, and some show FPS losses.
Blur Busters forumsWhy we don't: it removes boost headroom. Custom-OS projects added guards against it after users lost performance.
AtlasOS 0.4.0 release notesGPU Priority / SFIO PriorityWhy we don't: Microsoft documents these values as not used, and in testing the Games task has never been observed in use by games.
Microsoft Learn: MMCSS GamingPCSetup researchSystemResponsiveness = 0Why we don't: Windows treats any value below 10 as the default of 20, so setting 0 changes nothing.
Microsoft Learn: MMCSSNetworkThrottlingIndex = 0xFFFFFFFFWhy we don't: it measured higher network DPC latency than the default.
GamingPCSetup network researchIRQ8Priority, SecondLevelDataCache, IoPageLockLimit, LargeSystemCacheWhy we don't: folklore or server settings, with no evidence they help games on modern Windows.
PC-Tuning researchWin32PrioritySeparation values above 0x3FWhy we don't: the value wraps around. There's no special hidden mode.
PC-Tuning researchMouseDataQueueSize / KeyboardDataQueueSizeWhy we don't: these only set the size of an input buffer (100 entries by default). Smaller values risk dropped input, with no latency gain.
GamingPCSetup driver research RepairBadTweaksThreadDpcEnable = 0Why we don't: it turns threaded DPCs back into ordinary DPCs, which raises latency.
RepairBadTweaksTdrLevel = 0 or changing TdrDelayWhy we don't: these control how Windows recovers from a GPU hang, and Microsoft says end users shouldn't change them.
Microsoft Learn: TDR registry keysRMHdcpKeyglobZero)Why we don't: it breaks protected video playback, driver updates revert it, and there's no latency data behind it.
GamingPCSetup driver researchWhy we don't: borderless games running in independent flip mode match exclusive fullscreen.
Blur Busters forumsWhy we don't: at the same display mode the difference is under 1 ms, and DSC measured 4.714 ms on versus 4.718 ms off.
Blur Busters forums JoltflyWhy we don't: it saves at most about 0.875 ms compared with 1000 Hz, and it adds CPU load.
Tech4GamersWhy we don't: it helps only when the GPU is near full load. Otherwise it can add latency.
Beyond3D forumWhy we don't: it briefly raises CPU clocks for Start, search and app launches, and it has been on by default since June 2026.
Windows LatestWhy we don't: the 20% is a cap on priority traffic. It isn't bandwidth held back from you.
How-To GeekWhy we don't: Microsoft advises against it. It can cause delays and breaks some Windows features.
Microsoft Learn: IPv6 guidanceWhy we don't: it breaks Xbox networking and party chat.
Why we don't: DNS only affects how fast names are looked up. It doesn't change the latency of a game connection that's already established.
Why we don't: games use UDP, so turning autotuning off only caps TCP download speed.
Microsoft Learn: NIC tuning TweakHoundWhy we don't: most routers pass the tag along without honoring it. It only helps on your own router or Wi-Fi.
ScienceDirect study RFC 8325Why we don't: Off raised interrupts to about 2.65 million per second, and Medium measured as well or better.
GamingPCSetup network researchWhy we don't: above roughly 30 Mbps the benefit is marginal.
NetForecast report (PDF)Why we don't: relay paths can beat direct routes, but only when your ISP's routing is poor. NOVA's route monitor shows whether yours is.
arXiv studyWhy we don't: the internet carries 1500-byte packets, so jumbo frames don't help online games.
-high in CS2Why we don't: there's no measurable gain on modern systems, and it can compete with audio and background threads.
swap.gg CSMarketCap-threads N in CS2 or Dota 2Why we don't: Source 2 detects your cores itself, and forcing a count is linked to stutter and crashes.
swap.gg CSMarketCap MissOutPC-tickrate 128 in CS2Why we don't: it only affects local or self-hosted servers. Matchmaking uses sub-tick.
CSMarketCap-d3d9ex, -nod3d9ex, -dxlevel, -32bit, -tcpWhy we don't: they're Source 1 and DirectX 9 leftovers that do nothing in CS2 or Dota 2.
swap.gg MissOutPCWhy we don't: a Valve developer has said the default launch options are the most heavily tested (originally about CS:GO).
Steam Community (Valve developer)Why we don't: it blocks legitimate patch updates to settings and hides resets.
Why we don't: each one opens a security hole, and they also break anti-cheat.
PC-TuningWhy we don't: it weakens protection for no measured gain.
Why we don't: Windows Defender now flags WinRing0 as a vulnerable driver.
PCWorldWhy we don't: games that reserve a lot of memory crash without one.
Why we don't: there's no gain on SSDs, and maintenance is needed for drive optimization.
Why we don't: Game Mode helps when there's background load, and DLSS Frame Generation requires hardware-accelerated GPU scheduling.
csrss via IFEOWhy we don't: it can starve input and audio.
Why we don't: they can break Windows Update, the Microsoft Store and Xbox features for gains at noise level.
Frozen Tweaks guideWhy we don't: it breaks AMD's core parking, which keeps games on the V-Cache chiplet.
Hardware BustersWhy we don't: independent tests found a median gain of about +2 FPS, within the margin of error, and some apps lost FPS.
BetterFPS testWhy we don't: Microsoft does not support them.
Microsoft support policyWhy we don't: a product's own published benchmark tops out at 28.2%, partly from lowered in-game settings.
igor'sLAB Hardware BustersWhy we don't: results depend on the game, and testing across many games says to keep E-cores on.
TechPowerUp (SMT test)Every tweak carries an evidence rating that says how well it's proven and where the evidence comes from. Every change is backed up and reverts in one click. Proof mode repeats a test before and after a change and tells you whether it made things better, worse or made no measurable change. And NOVA installs no kernel drivers of its own for tuning or sensors.