Windows 11 hack: Higher SSD speeds with new Microsoft NVMe driver

[Redaktion]

It only takes three registry entries to get Windows 11 to use the native NVMe driver that was recently introduced with a Windows Server 2025 update, which can significantly increase SSD performance, as we have just tested. Those who do not mind a little risk can hit data transfer rates that are sometimes up to 45% higher for exactly zero dollars.

https://www.notebookcheck.net/Windows-11-hack-Higher-SSD-speeds-with-new-Microsoft-NVMe-driver.1190489.0.html

[Im not ai]

Please wake me up when they resurrect windows 95

[vr]

This does nothing. Reg entries added and nothing changes. Maybe it's just me?

[evol]

Win 11 24H2, Samsung 970 evo 512, samsung 980 512 and Fikwot FN501 Pro 1tb working in PCI-ex 3.0 adapter worked great. Random access improved almost 2x for RND4K Q32T1! Chinese cheapest SSD with new driver outperforms Samsung 970 EVO with old driver. Insane upgrade! Great job, Microsoft /s

[Jay F]

Don't bother with this.

Registry additions and driver changes all came through for me after the edits with no issues.

Not only did it have no effect on the r/w for my secondary SSD, it actually tested worse in both by 200MB/s in CrystalDiskMark. Acer Predator GM7000 4TB, PCIe4, on a Gigabyte G5 KD.

Again, don't bother.

[devdd]

This does nothing. Reg entries added and nothing changes. Maybe it's just me?

me too. Not working.

[JB996]

One test was slower, but overall there were some serious increases.
I'm not debating it, just presenting my results.

Before my SSDs were 'Disk Drives' in Device Manager.
After reg addition and restart, they show up as Storage Disks.

CrystalDiskMark 9.0.1 x64
3x 1GiB

Samsung SSD 980 PRO 500GB
......................................Before     After (MB/s)
SEQ1M Q8T1 Read   6821.22    6744.65
SEQ1M Q8T1 Write  4721.48    4874.89
SEQ1M Q1T1 Read   4175.48    4192.28
SEQ1M Q1T1 Write  3046.16    4296.56     
RND4k Q32T1 Read  757.88     1033.88
RND4k Q32T1 Write 662.22     866.46
RND4k Q1T1 Read   92.92      92.43
RND4k Q1T1 Write  239.65     255.22

Samsung SSD 990 PRO with Heatsink 4TB
......................................Before     After (MB/s)
SEQ1M Q8T1 Read   7044.36    7129.50
SEQ1M Q8T1 Write  6626.18    6967.53
SEQ1M Q1T1 Read   3922.78    3333.19
SEQ1M Q1T1 Write  5616.99    5743.89
RND4k Q32T1 Read  808.83     1114.92
RND4k Q32T1 Write 694.93     932.09
RND4k Q1T1 Read   80.37      87.73
RND4k Q1T1 Write  253.24     278.35

[Jon Pussi]

Dear Writer,

Thanks!! It did work on my 25H2!!! I see the Storage Disk on device manager now Hoooray!!!!

[Bud]

The system cannot find the path specified. 25H2   26200.7462

[Sylvain]

I see huge gains in Q32T1: +40% reading, +60% writing

[Frageok]

I did the hack in a system with 3 NVME SSDs installed. Brands are Lexar, Fikwot and Samsung. The registry hack worked and the system feels snappier.

[Andi Kassner]

Nach dem ich diesen "Hack" ausprobierte musste ich ihn wieder entfernen weil Acronis True Image die SSDs nicht mehr anzeigte bzw. als interne Laufwerke erkannte.
Bei Tests mit AS SSD Benchmark hatte ich kaum Unterschiede festgestellt.
Entfernt habe ich die neuen Schlüssel mit REGEDIT und den neu angelegten Ordner
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft gelöscht indem ja sowieso nur Daten für diesen Hack  enthalten sind. Nach einem Neustart war alles wieder wie vorher - ich befürchtete das Schlimmste -  und Acronis hat auch wieder meine internen Laufwerke erkannt.
Am besten wäre es, man löscht hier diesen Artikel weil es auch viele Benutzer mit wenig/keiner Erfahrung gibt um das dann wieder zurück zu nehmen. Meiner Meinung nach sollte dieser "Hack" bei Server 2025 bleiben wo es auch Sinn macht.

[anon718381]

I decided to test this hack thoroughly on my rig. Here are my results using CrystalDiskMark (Settings: NVMe Profile, 5 Passes, 4GiB size to properly stress the cache).

My Setup:
Drive: WD_BLACK SN850X 2000GB (NVMe Gen4).
Condition: Firmware updated to the latest version (620361WD.

The Results: I compared the "Server 2025 Driver" (Hack) against the Stock Windows Driver.

Contrary to the article, the hack provided zero benefit to Sequential speeds on my drive. However, the hack severely impacted Random 4K performance (latency and daily responsiveness).

Metric    Registry Hack (Server Driver)    Stock Driver
Seq Read (Q8T1)    7,365 MB/s    7,365 MB/s
Identical

Seq Write (Q8T1)    6,614 MB/s    6,595 MB/s
Identical
RND 4K Read (Q1T1)    65.77 MB/s    74.68 MB/s
Hack is -12% Slower

RND 4K Write (Q1T1)    159.36 MB/s    250.84 MB/s
Hack is -36% Slower

Conclusion: For a high-end consumer drive like the SN850X, this hack degrades performance where it matters most (Random 4K). The "Server" driver seems to prioritize queue management in a way that hurts low-latency consumer tasks. I strongly recommend sticking with the stock driver.

[elstak]

It's clearly indicated 25H2, why ppl are using it on 24H2 is why the world is goning to hell