Microsoft Unlocks Native NVMe Support in Windows Server 2025, Delivering 80% IOPS Boost

Delivering on performance promises made during its November launch, Microsoft has formally released Native Non-Volatile Memory Express (NVMe) support for Windows Server 2025. This architectural overhaul unlocks an 80% gain in input/output operations per second (IOPS), bypassing legacy bottlenecks.

Despite reaching General Availability (GA) in the October cumulative update, the feature remains disabled by default. Administrators must manually activate the new stack via Registry or Group Policy changes to bypass the legacy Small Computer System Interface (SCSI) translation layer.

Breaking the SCSI Bottleneck

For decades, Windows Server has treated all storage devices through a unified lens, forcing even modern flash memory to communicate via the SCSI protocol. While effective for rotational hard drives, this translation layer became a significant choke point for high-speed solid-state media.

Under the hood, the architectural changes represent a fundamental break from this legacy model. By eliminating the translation step, the new Native NVMe stack aligns the operating system with the parallel capabilities of modern hardware.

Yash Shekar, a Program Manager at Microsoft, noted that “this improvement comes from a redesigned Windows storage stack that no longer treats all storage devices as SCSI devices, a method traditionally used for older, slower drives.”

Technically, the disparity between the two approaches is substantial. Legacy SCSI stacks are limited to a single queue capable of handling just 32 concurrent commands. In contrast, the modern NVMe protocol was designed for high parallelism, supporting up to 64,000 queues with 64,000 commands each.

Microsoft’s internal benchmarking highlights the practical impact of this shift. Tests conducted on an Intel Dual Socket system with 128GB of RAM and a Solidigm NVMe drive revealed an 80% increase in IOPS performance.

Concurrently, the benchmarks showed a 45% reduction in CPU cycles per I/O operation, freeing up compute resources for application workloads. The announcement details the specific limitations of the legacy architecture:

“SCSI-based I/O processing can’t keep up because it uses a single-queue model, originally designed for rotational disks, where protocols like SATA support just one queue with up to 32 commands.”

These architectural limitations have long forced high-performance deployments to rely on vendor-specific drivers or specialized hardware configurations to bypass the OS bottleneck. Ultimately, the native stack aims to democratize this performance on standard hardware.

The ‘Opt-In’ Catch: How to Enable Native NVMe

Although the technology is now technically Generally Available (GA) following the launch of Windows Server 2025, Microsoft has adopted a conservative deployment strategy. Included in the October 2025 Cumulative Update (KB5066835), the feature remains inactive out of the box.

Administrators looking to utilize the new stack must manually intervene. Microsoft is encouraging users to switch, warning that failing to do so leaves significant performance on the table.

Shekar emphasized the urgency for enterprise environments, stating: “Switch onto Native NVMe as soon as possible or you are leaving performance gains on the table!”

Activation requires a specific modification to the Windows Registry or the application of a Group Policy Object (GPO). Such an opt-in model suggests Microsoft is prioritizing stability, allowing organizations to validate the new storage path in staging environments before broad deployment.

To enable the feature, administrators can execute the following PowerShell command:

reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides /v 1176759950 /t REG_DWORD /d 1 /f

 
A critical prerequisite for this activation is the use of the in-box StorNVMe.sys driver. Systems configured with third-party vendor drivers will continue to use those specific paths and will not trigger the native Windows stack improvements.

Deployment friction here mirrors other recent administrative hurdles, such as the previous auto-upgrade errors that inadvertently pushed the new OS to some servers, and changes to hotpatching subscription changes that altered cost structures for enterprise management.

Real-World Impact: From SQL to AI

For administrators managing high-density virtualization clusters, the efficiency gains offer tangible economic benefits. Lower CPU overhead per storage operation allows for higher virtual machine density on the same physical host, potentially reducing hardware procurement needs.

Microsoft has identified several key workloads that stand to benefit most from the upgrade. Transaction-heavy SQL Server environments (OLTP) are expected to see reduced latency, while Hyper-V deployments should experience faster live migrations.

Data-intensive operations also see improvement. AI and machine learning workflows, which often involve shuffling large datasets between storage and memory, can leverage the increased throughput to reduce training and inference times.

Shekar explained the mechanism behind these latency improvements, emphasizing that “Native NVMe enables streamlined, lock-free I/O paths that slash round-trip times for every operation.”

External validation has already begun to corroborate Microsoft’s internal figures. Storage vendor StarWind conducted independent testing of the new stack, achieving 3.4 million IOPS on commodity hardware.

This third-party data supports the claim that the bottleneck has effectively shifted from the software stack back to the hardware itself. By removing the software limitations, Windows Server 2025 is now positioned to fully utilize the speed of PCIe Gen5 storage devices.

Shekar concluded that “direct, multi-queue access to NVMe devices means you can finally reach the true limits of your hardware.”