winjer
Member
Announcing Native NVMe in Windows Server 2025: Ushering in a New Era of Storage Performance | Microsoft Community Hub
We’re thrilled to announce the arrival of Native NVMe support in Windows Server 2025—a leap forward in storage innovation that will redefine what’s possible...
Modern NVMe devices—like PCIe Gen5 enterprise SSDs capable of 3.3 million IOPS, or HBAs delivering over 10 million IOPS on a single disk—are pushing the boundaries of what storage can do. 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. In contrast, NVMe was designed from the ground up for flash storage and supports up to 64,000 queues, with each queue capable of handling up to 64,000 commands simultaneously.
With Native NVMe in Windows Server 2025, the storage stack is purpose-built for modern hardware—eliminating translation layers and legacy constraints. Here's what that means for you:
- Massive IOPS Gains: Direct, multi-queue access to NVMe devices means you can finally reach the true limits of your hardware.
- Lower Latency: Traditional SCSI-based stacks rely on shared locks and synchronization mechanisms in the kernel I/O path to manage resources. Native NVMe enables streamlined, lock-free I/O paths that slash round-trip times for every operation.
- CPU Efficiency: A leaner, optimized stack frees up compute for your workloads instead of storage overhead.
- Future-Ready Features: Native support for advanced NVMe capabilities like multi-queue and direct submission ensures you're ready for next-gen storage innovation.
Using DiskSpd.exe, basic performance testing shows that with Native NVMe enabled, WS2025 systems can deliver up to ~80% more IOPS and a ~45% savings in CPU cycles per I/O on 4K random read workloads on NTFS volumes when compared to WS2022. This test ran on a host with Intel Dual Socket CPU (208 logical processors, 128GB RAM) and a Solidigm SB5PH27X038T 3.5TB NVMe device. The test can be recreated by running "diskspd.exe -b4k -r -Su –t8 -L -o32 -W10 -d30" and modifying the parameters as desired. Results may vary.
To enable the feature. It also works on Windows Pro and Home. Though this is still a feature only meant for servers.
But please make sure to backup your registry first, in case something goes wrong.
It requires KB5066835 to be installed. But this was released several months ago, and most people should have it.
It also requires Windows 11 24H2 or 25H2.
reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides /v 1853569164 /t REG_DWORD /d 1 /f
reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides /v 156965516 /t REG_DWORD /d 1 /f
reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides /v 735209102 /t REG_DWORD /d 1 /f
My experience:
I didn't do proper benchmarks, just ran a few games to see if there was any difference.
With games that use Direct Storage, such as BF6, Spiderman I didn't notice any changes. With Ratchet and Clank the loading seemed faster, but could be just my impression.
But then I tested games using UE4 and 5. One thing that always happened wit these games is that due to it's streaming system, when I started running in game, performance would drop 10-15 FPS, and CPU utilization would increase a bit.
So I installed Kena, The Forbidden City, High on Life, and Talos Principle Reawakened. These are games where I had previously noticed they would drop performance whenever I started running in game.
With this driver stack, CPU utilization barely moved. And the frame rate was almost stable, only changing a bit due to the scenery moving. But not the drop in performance that UE4 and 5 games usually have.
Also, stutters were fewer. Talos Principle always ran very well. But Kena and The Forbidden City and High on Life, always had a good amount of stutters, even after shader compilation, due to asset streaming.