Microsoft’s latest native NVMe driver, initially rolled out for Windows Server 2025, is now making waves in the consumer space with Windows 11. This driver promises to enhance the performance of solid-state drives (SSDs), particularly for those who are keen on optimizing their storage solutions. With a few simple registry tweaks, users can unlock impressive performance gains that have been highlighted by StorageReview’s extensive testing.
Microsoft Native NVMe Driver Performance
The native NVMe driver has shown significant improvements across three critical areas of storage performance. The most notable enhancements are in random read bandwidth and input/output operations per second (IOPS). These advancements lead to quicker data access and more efficient operations, especially when systems are under heavy loads or multitasking.
In a recent benchmark test featuring AMD EPYC 9754 processors, 768GB of DDR5-4800 memory, and 16 Solidigm P5316 30.72TB PCIe 4.0 SSDs configured in a JBOD setup, the results were compelling:
<table tabindex="0" class="tablewrapper tablewrapper–inbodyContent tablewrapper–sticky tablewrapper–divider”>
<thead class="tablehead”>
<tr class="tableheadrow”>
<th class="tableheadheading tableheadheading–left” colspan=”1″>
<th class="tableheadheading tableheadheading–left” colspan=”1″>Non-Native Driver
<th class="tableheadheading tableheadheading–left” colspan=”1″>Native Driver
<th class="tableheadheading tableheadheading–left” colspan=”1″>Improvement
<tbody class="tablebody”>
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>4K Random Read (GiB/s)
<td class="tablebody_data” colspan=”1″>6.1
<td class="tablebody_data” colspan=”1″>10.058
<td class="tablebody_data” colspan=”1″>+64.89%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>64K Random Read (GiB/s)
<td class="tablebody_data” colspan=”1″>74.291
<td class="tablebody_data” colspan=”1″>91.165
<td class="tablebody_data” colspan=”1″>+22.71%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>64K Sequential Read (GiB/s)
<td class="tablebody_data” colspan=”1″>35.596
<td class="tablebody_data” colspan=”1″>35.623
<td class="tablebody_data” colspan=”1″>+0.08%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>128K Sequential Read (GiB/s)
<td class="tablebody_data” colspan=”1″>86.791
<td class="tablebody_data” colspan=”1″>92.562
<td class="tablebody_data” colspan=”1″>+6.65%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>64K Sequential Write (GiB/s)
<td class="tablebody_data” colspan=”1″>44.67
<td class="tablebody_data” colspan=”1″>50.087
<td class="tablebody_data” colspan=”1″>+12.13%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>128K Sequential Write (GiB/s)
<td class="tablebody_data” colspan=”1″>50.477
<td class="tablebody_data” colspan=”1″>50.079
<td class="tablebody_data” colspan=”1″>-0.79%
According to the benchmarks, random read performance experienced the most substantial improvements, with 4K and 64K read speeds increasing by 64.89% and 22.71%, respectively. Sequential read performance, particularly at the 64K block size, remained stable, while a shift to 128K yielded a modest 6.65% boost. In terms of sequential write performance, the 64K block size showed a commendable 12.13% increase, although the 128K block size did not provide any further advantage.
Latency testing revealed a mixed bag of results. Random read latency saw significant reductions, with 4K and 64K read times decreasing by 38.46% and 13.39%, respectively. Conversely, sequential write latency increased, particularly at the 64K block size, which rose by 39.85%. However, switching to a 128K block size mitigated this increase, resulting in a more manageable 12.43% rise.
<table tabindex="0" class="tablewrapper tablewrapper–inbodyContent tablewrapper–sticky tablewrapper–divider”>
<thead class="tablehead”>
<tr class="tableheadrow”>
<th class="tableheadheading tableheadheading–left” colspan=”1″>
<th class="tableheadheading tableheadheading–left” colspan=”1″>Non-Native Driver
<th class="tableheadheading tableheadheading–left” colspan=”1″>Native Driver
<th class="tableheadheading tableheadheading–left” colspan=”1″>Improvement
<tbody class="tablebody”>
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>4K Random Read Latency (ms)
<td class="tablebody_data” colspan=”1″>0.169
<td class="tablebody_data” colspan=”1″>0.104
<td class="tablebody_data” colspan=”1″>-38.46%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>64K Random Read Latency (ms)
<td class="tablebody_data” colspan=”1″>0.239
<td class="tablebody_data” colspan=”1″>0.207
<td class="tablebody_data” colspan=”1″>-13.39%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>64K Sequential Write Latency (ms)
<td class="tablebody_data” colspan=”1″>0.399
<td class="tablebody_data” colspan=”1″>0.558
<td class="tablebody_data” colspan=”1″>+39.85%
<tr class="tablebodyrow”>
<td class="tablebody_data” colspan=”1″>128K Sequential Write Latency (ms)
<td class="tablebody_data” colspan=”1″>1.022
<td class="tablebody_data” colspan=”1″>1.149
<td class="tablebody_data” colspan=”1″>+12.43%
In terms of CPU usage, the NVMe driver also demonstrated favorable outcomes across both sequential read and write operations. Notably, the driver is not enabled by default in Windows 11; users must opt in by making specific registry adjustments. This decision reflects Microsoft’s commitment to ensuring compatibility and support from third-party vendors before fully integrating the native NVMe driver into the operating system.
