Windows Server admins finally have a reason to get excited about storage plumbing.
Microsoft says Windows Server 2025 now supports native NVMe I/O—meaning the OS no longer has to convert NVMe commands into SCSI commands just to push I/O through the storage stack. The change has reached General Availability starting with Windows Server 2025 and the latest October cumulative update, but it’s opt-in and disabled by default.
What “native NVMe I/O” changes
Historically, Windows has treated storage devices through a SCSI-oriented path, even when the underlying device is NVMe. With native NVMe enabled, Microsoft says it has redesigned the I/O processing workflow for extreme performance, including more streamlined, lock-free paths intended to reduce round-trip times under load.
This isn’t a feature you install as an add-on. It’s built into the OS—your job is deciding whether to enable it.
How to enable it
Microsoft says enabling native NVMe I/O can be done via registry or Group Policy. The registry method is straightforward:
reg add HKLM\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides ^
/v 1176759950 /t REG_DWORD /d 1 /f
Microsoft also provides a Group Policy MSI approach for managed environments.
One important caveat: the benefits are tied to the in-box Windows NVMe driver (StorNVMe.sys). If a system is using a vendor-supplied NVMe driver, results may vary—and some environments may see little to no change.
Performance claims: big gains, but from a specific test
Microsoft’s headline numbers are attention-grabbing: up to ~80% higher IOPS and up to ~45% fewer CPU cycles per I/O in high-I/O scenarios.
Those figures come from a specific benchmark setup (DiskSpd 4K random reads on NTFS) on a dual-socket Intel server-class system with 208 logical cores, 128GB RAM, and a Solidigm 3.5TB PCIe 5.0 enterprise SSD (published by part number, commonly associated with the D7-PS1010 family).
Microsoft’s charts also suggest thread scaling improvements—e.g., meaningful gains even with a single I/O thread and strong jumps at higher thread counts—but those per-thread percentages are best treated as chart-derived rather than universal expectations. Your workload, queue depth, filesystem behavior, and driver stack will heavily influence outcomes.
Who should care
If the numbers hold up in real deployments, native NVMe I/O is most likely to matter where storage becomes the bottleneck:
- High-performance file serving
- Virtualization hosts
- Database-heavy systems
- AI/ML pipelines that thrash local NVMe
- Any environment where CPU overhead per I/O is a limiting factor
What about Windows 11?
For now, Microsoft hasn’t announced a timeline for bringing this same native NVMe I/O path to “plain” Windows 11.
It’s tempting to assume it will show up and instantly improve gaming, DirectStorage, or “general snappiness,” but that’s still speculation. Consumer SSD firmware, driver differences, and the wildly varied I/O patterns on desktops make this the kind of feature Microsoft will likely need to validate carefully before rolling it out broadly—and it may end up being enabled selectively depending on the device and driver in use.
The bottom line
Windows Server 2025’s native NVMe I/O is a rare kind of server feature: it’s deep in the stack, boring on the surface, and potentially huge in practice. If you’re running high-I/O systems and you’re already on Server 2025 with the latest October cumulative update, it’s a feature worth lab testing—especially if you’re chasing more IOPS without paying the CPU tax.