We took one machine, a stopwatch and a frame-time logger, and worked through every popular Windows 11 “speed tweak” we could find. Most of them did nothing measurable. Two dozen did. These are the two dozen, with the numbers.
There is an entire cottage industry built on Windows optimisation, and almost none of it publishes measurements. Registry “cleaners”, service-disabling scripts, mystery .bat files with four hundred lines — the pattern is always the same: a confident claim, no before-and-after, and a placebo effect strong enough to keep the genre alive for thirty years.
So we did the boring thing. One test machine, one variable at a time, five runs per configuration, first run discarded. Everything below moved a number we could reproduce. Everything in the myths section did not.
How we measured, and why most tweaks fail
Our test system is deliberately mid-range, because that is where optimisation has room to matter: an eight-core processor, 32 GB of DDR5, a 2 TB PCIe 4.0 SSD and a mid-range graphics card, on a clean Windows 11 install with current drivers. Four measurements, repeated after every change:
- Cold boot to interactive desktop — stopwatch from power button to the moment a click on the Start button responds, not to the moment the wallpaper appears.
- Application cold-launch time — the average across five heavy applications after a reboot.
- 1% low frame times in a CPU-bound game scene, because averages hide stutter.
- Idle background CPU and disk activity over ten minutes with nothing running.
Our measured variance across five runs was ±1.4% on frame rate and ±0.8 s on cold boot. Any “improvement” smaller than that is noise, and a great deal of published optimisation advice lives entirely inside that band.
The reason most tweaks fail is simple and unglamorous: Windows 11 already does the thing. Memory compression, prefetching, service triggers, scheduled defragmentation with SSD-aware TRIM — the operating system has spent fifteen years absorbing what used to require manual intervention. What is left is a much smaller set of decisions, mostly about what you have installed and how you have configured your hardware.
Startup and background load (settings 1–5)
This category produced our single largest measured win, and it is entirely unglamorous.
1. Prune startup applications properly
Task Manager → Startup apps sorts by “Startup impact”, which is a rough guide at best. The better signal is what a program actually does at boot. Updaters, cloud-sync clients and vendor “assistants” are the usual culprits. On our deliberately dirty test image, disabling eleven startup entries took cold boot from 42.1 s to 24.8 s — a 41% improvement that no registry tweak came close to.
2. Disable startup tasks in Task Scheduler, not just Startup apps
Many applications register a scheduled task that runs at logon, which does not appear in the Startup list at all. Open Task Scheduler, look under Task Scheduler Library and its vendor sub-folders, and disable anything triggered “At log on” that you do not need. Worth a further 2–4 seconds on our image.
3. Turn off “Restart apps after sign-in”
Settings → Accounts → Sign-in options. Windows reopens whatever was running when you shut down. Convenient occasionally, expensive every time. 3.1 s off our cold boot.
4. Remove the second antivirus
Running a third-party suite alongside Microsoft Defender does not double your protection; it doubles your file-system filter drivers. Every file open is scanned twice by two drivers that do not know about each other. In our test, a second real-time scanner added 18% to application cold-launch times and produced visible stutter during game asset streaming. Pick one.
5. Uninstall rather than disable
A disabled service that is still installed still has its driver loaded. If you will never use a piece of vendor software, remove it. This is the least exciting advice in the article and reliably the most effective.
If you do exactly one thing from this article, do setting 1. Startup pruning beat every other single change we measured, on every machine we tried it on, and it takes four minutes.
Power, scheduling and latency (6–10)
6. Set the power mode to Best Performance on desktops
Settings → System → Power. On a desktop this raises the processor's minimum state and shortens the ramp-up delay. Measured effect on average frame rate: +1.9%. Measured effect on 1% lows: +7.4%, because the scheduler stops dropping to low clocks between frames. On a laptop, do this only when plugged in — the battery cost is real.
7. Enable Hardware-Accelerated GPU Scheduling
Settings → System → Display → Graphics → Change default graphics settings. Moves some scheduling work from the CPU to the GPU's own scheduler. On our system: +3.2% average frame rate, −11% frame-time variance. Not universal — on two older GPUs we tested it made things marginally worse, so measure rather than assume.
8. Turn off “Enhance pointer precision”
This is mouse acceleration. It does not change frame rate but it changes how the machine feels, and in a shooter it is the difference between muscle memory working and not working. Settings → Bluetooth & devices → Mouse → Additional mouse settings → Pointer Options.
9. Set the correct refresh rate
Genuinely common: a 144 Hz monitor running at 60 Hz because nobody changed it. Settings → System → Display → Advanced display. Check every monitor separately; Windows does not apply this across displays.
10. Disable full-screen optimisations selectively
Right-click an executable → Properties → Compatibility. This forces true exclusive full-screen rather than a borderless composited window. Results were genuinely mixed: −4 ms input latency in three older titles, no measurable change in five modern ones, and one title that started stuttering on alt-tab. Try it per-game, keep it where it helps.
Settings 6 and 7 trade battery life for responsiveness. On a laptop, configure Best Performance for the plugged-in profile only and leave the battery profile balanced. A laptop that is fast for ninety minutes is not fast.
Storage and memory behaviour (11–15)
11. Keep 20–25% of your SSD free
The largest single storage effect we measured, by a wide margin. As an SSD fills, the controller has fewer free blocks to write into and must erase before writing. We filled the same drive progressively and measured sustained write throughput:
| Drive utilisation | Sustained write | Relative |
|---|---|---|
| 40% full | 4,910 MB/s | 100% |
| 70% full | 4,640 MB/s | 94% |
| 85% full | 2,880 MB/s | 59% |
| 95% full | 1,120 MB/s | 23% |
| 99% full | 410 MB/s | 8% |
12. Leave the page file alone
“Disable the page file, you have 32 GB of RAM” is confidently wrong. Windows uses the page file for memory it has decided is cold, and some applications crash outright without one. Our test with the page file disabled produced two application crashes and no measurable performance gain. System-managed is correct.
13. Verify Storage Sense is on, and configured
Settings → System → Storage. It clears temporary files and empties the recycle bin on a schedule. Combined with setting 11, this is how a drive stays out of the performance cliff without you thinking about it.
14. Check that TRIM is actually running
Open an elevated prompt and run fsutil behavior query DisableDeleteNotify. A result of 0 means TRIM is enabled. Third-party “optimiser” tools sometimes disable it while claiming the opposite.
15. Move the shader cache off a full drive
Graphics driver shader caches can reach several gigabytes and are written constantly. On a machine with a small, full boot drive this is a real source of stutter. Both major drivers allow relocating the cache; do it if your boot drive is under 500 GB.
It is the single most repeated piece of bad Windows advice. Windows does not use the page file the way Windows XP did. Disabling it gains nothing measurable and causes hard crashes in applications that map large files. System-managed, always.
Graphics and display stack (16–19)
16. Do a clean driver installation, not an upgrade
Layered driver installations accumulate stale registry entries and orphaned control-panel profiles. A clean install — using the driver package's own clean option or a dedicated removal tool in Safe Mode — resolved a persistent 1% low regression on one of our test machines that three “upgrade” installs had not touched. Recovered 9% on 1% lows.
17. Turn off desktop transparency effects
Settings → Personalisation → Colours. Marginal on a modern GPU, genuinely noticeable on integrated graphics: +2.8% average frame rate on our integrated-graphics laptop, effectively zero on the desktop.
18. Disable animation effects on low-end hardware
Settings → Accessibility → Visual effects. The interface feels faster because it is faster to draw, not because you are imagining it. On the laptop, window-open latency dropped from 180 ms to 95 ms.
19. Set per-application GPU preference on laptops
Settings → System → Display → Graphics. Hybrid laptops sometimes run a game on the integrated GPU. Explicitly assigning High Performance to the executable is a two-click fix for a problem that otherwise looks like a broken graphics card.
Security features that cost performance (20–22)
The three settings in this section trade real security for a few per cent of performance. That can be a defensible choice on a dedicated gaming machine that holds nothing valuable. It is not a defensible choice on a machine you bank, work or store client data on. Decide deliberately, and write down what you changed so you can reverse it.
20. Memory Integrity (HVCI)
Windows Security → Device security → Core isolation. This runs the kernel under a hypervisor to block unsigned driver code. It is a genuine security control, and it costs performance: we measured −4.1% average frame rate and −9.3% on 1% lows. On a gaming machine that never sees untrusted drivers you may decide that trade is worth it. On a work machine, leave it on. Do not disable it because a video told you to — decide deliberately.
21. Virtual Machine Platform
If you do not use WSL, Android emulation or Hyper-V, this feature is enabling hypervisor overhead for nothing. Turning it off recovered 2.2% average frame rate on our system. If you do use those things, keep it — the overhead is the price of the feature.
22. Exclude game directories from real-time scanning
Windows Security → Virus & threat protection → Exclusions. Scanning thousands of asset files as a game streams them produces measurable stutter. Excluding only your game library folder cut frame-time spikes by 14% in one open-world title. Exclude specific folders, never whole drives, and never your downloads folder.
Maintenance that genuinely matters (23–24)
23. Update firmware, not just drivers
Motherboard firmware, SSD firmware and monitor firmware all ship fixes that affect stability and performance. An SSD firmware update on one test drive resolved a latency spike that had been showing up in our logs for months as random one-second freezes. Check twice a year; firmware updates are not like driver updates, they are rare and usually meaningful.
24. Reinstall once every two or three years
Unfashionable advice, still true. Windows accumulates driver filters, shell extensions, orphaned services and installer detritus that no cleaning tool reliably removes. Our aged test image, after every optimisation above, was still 19% slower on cold boot than a fresh install of the same build on the same hardware. Back up, reinstall, and copy your files back rather than restoring an image.
Export your firmware settings, note your memory profile, list your installed applications and copy your saved game folders. The reinstall itself takes twenty minutes; reconstructing what you forgot takes a fortnight.
Eleven tweaks that did nothing at all
Every one of these was measured with the same methodology. Every one produced a difference inside our run-to-run variance, which is to say: nothing.
| Popular tweak | Claim | Measured result |
|---|---|---|
| Registry cleaning | “Faster boot, fewer errors” | No change. Some tools broke file associations. |
| Disabling SysMain / Superfetch | “Less disk activity” | No change on SSD; slower app launches on HDD. |
| “Ultimate Performance” power plan | “Unlocks hidden speed” | Identical to Best Performance within variance. |
| Disabling Windows Search indexing | “Frees CPU” | No change; search became useless. |
| Setting processor affinity per game | “Dedicated cores” | No change, occasionally worse on hybrid CPUs. |
| Increasing “processor cores” in msconfig | “Uses all cores at boot” | No change. The setting only ever limits cores. |
| Disabling the Print Spooler | “One less service” | No change. Printing stopped working. |
| Third-party RAM “optimisers” | “Frees memory” | Forced working sets out of RAM; measurably worse. |
| Disabling telemetry services | “Faster, more private” | No performance change. Privacy case is separate and legitimate. |
| Defragmenting an SSD manually | “Tidier data” | No gain, unnecessary write wear. |
| “Game booster” utilities | “Closes background tasks” | Within variance; one added its own overlay overhead. |
If a tweak is genuinely worth doing, someone can show you the before-and-after. If nobody can, you are looking at folklore with a download button. Mary R. Hargis, Editor-in-Chief
A twenty-minute quarterly routine
Doing all twenty-four settings once is worth less than doing six of them every quarter. This is the checklist we actually use:
- Check free space on every SSD. Below 25% free, delete or move something today.
- Review startup apps and scheduled logon tasks. New software adds itself quietly.
- Uninstall anything you have not opened in six months.
- Check for firmware updates — motherboard, SSD, monitor. Twice a year is enough.
- Clean the dust filters. A five-minute job that is worth more thermally than every software setting on this page combined.
- Re-run one benchmark and compare it to last quarter. A 10% regression is a signal, not a mood.
A plain text file with the date, what you changed and the benchmark number afterwards. It takes thirty seconds and it turns “I think it feels slower” into a question you can actually answer.
If your machine is slow in a way none of this explains, the cause is usually hardware: a failing drive, degraded thermal paste, a dying power supply or memory that has quietly dropped to JEDEC speeds. Our diagnostic service covers exactly that, and the initial pass is free. You can also work through the storage guide first — storage is the culprit more often than anything else.
Questions readers keep asking
They will make it as fast as it was when new, which is not the same thing. Settings 1, 2, 4, 11, 17 and 18 are the ones that help most on older and lower-powered machines. Beyond that, the honest answer is usually hardware: adding memory to reach 16 GB and replacing a hard disk with an SSD will transform a 2018 laptop far more than anything on this page.
It is a real reduction in security, not a cosmetic one — it exists to stop malicious or vulnerable drivers from running kernel code. On a machine used only for games, with no work data on it, many people accept that trade for the 4–9% we measured. On any machine you bank or work on, leave it enabled. And if a game's anti-cheat requires it, the decision is made for you.
Not manually. Windows runs scheduled optimisation that issues TRIM to SSDs and defragments mechanical drives, and it picks the right operation per drive type. Manually defragmenting an SSD gains nothing and consumes write endurance. Just confirm the scheduled task is enabled under Optimise Drives.
In our experience, four causes account for most cases: the memory profile was never enabled in firmware, the monitor is plugged into the motherboard instead of the graphics card, the machine shipped with a trial antivirus running alongside Defender, or the boot drive is a DRAM-less budget SSD. All four are checkable in under ten minutes.
We have not found one that beat doing the six things in the quarterly routine by hand. Several made things measurably worse by forcing working sets out of memory or disabling TRIM. If you want one tool, use Task Manager and Storage Sense — both ship with the operating system and both actually tell you what they are doing.
