Hmmm. The NVMe standard has existed since 2011, and Samsung released their first commercially-available drive with it in 2013. So Microsoft has had at least 12 years to make nvmedisk.sys the standard driver for these disks.
this post was submitted on 26 Dec 2025
132 points (91.8% liked)
Technology
78002 readers
2180 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
Although Windows has had awareness of the NVMe storage media protocol since Windows 8.1, it turns out that the stock Microsoft driver for NVMe devices, disk.sys, offers suboptimal performance. This driver dates back to 2006, and is part of Microsoft's oldest internal basic drivers. Disk.sys appears to treat NVMe devices like SCSI drives. Microsoft released a new native driver with a greater degree of awareness of NVMe with Windows 11 25H2 (client) and Windows 2025 (server) operating systems, called nvmedisk.sys. The easiest way to check if your drive is using the older driver would be to bring up Device Manager, collapse "Disk Drives," open the Properties of your drive, go to the Driver tab, and click on the "driver details" button.
Notebookcheck made a fascinating discovery that has the potential to unlock greater performance with your NVMe drives, if they are compatible. Apparently, nvmedisk.sys significantly improves performance, both in sequential and random workloads. Using this driver, however, is fraught with risks. Not all NVMe SSDs support it, and if incompatible, it could break Windows 11 boot. The publication put out a guide on how to get Windows 11 to use nvmedisk.sys. This involves changing three Windows Registry values. It would be a good idea to image or backup your data before you tinker with this, so you can perform a full image restore if it breaks Windows booting. The guide can be found in the source links below, use it at your own risk.
Using this driver, however, is fraught with risks. Not all NVMe SSDs support it, and if incompatible, it could break Windows 11 boot.
Probably why it isn't standard, especially since there's a driver that does work even if it's suboptimal.
And obviously, there's been no possible way to try loading the modern driver and if that fails, falling back to the legacy one.
This is once again Microsoft refusing to improve performance, because that doesn't directly increase profits.
That's fair. I'm certainly not one to defend msoft, nor do I really have the technical knowledge to rebut. Is it possible that 'trying' the driver as you suggested could damage the drive or corrupt data? Just wondering if there's a legitimate reason they wouldn't go for a seemingly easy win aside from being a generally dumb organization.
There's always the option of gathering device info first, then using the appropriate driver. Either the SSD is in a "known supported models" list, or it reports support for whatever feature the new driver needs.
It's technically possible that straight up trying an unsupported driver can cause physical damage, but this can be avoided by carefully selecting the driver. From MS pov, they'd have to extensively test this driver on a bunch of SSDs and configurations, but it would lead to a performance improvement.
Ah, any developer who suggested that probably got the same answer I get at work: "Testing costs money, so unless we absolutely have to, no."
Is this not the test? Make it available, but you have to jump through hoops to enable?
If only there was a way to do a check for compatibility on the os side for a standard that has been available since before the predecessor os was released and fall back to the older driver if it fails
So they can't just write some probe code? It really can't be that hard to determine if there's support.
I'd be wary of importing random unintelligible registry keys from sites purporting to increase system performance for free. Even if they're posted on social media, the land of rigorous fact checking, rationality and absolutely no malicious actors.
It mentions & includes a link to the NotebookCheck site as the source. It's certainly NOT an unreliable site.
laughs in Linux
It's probably vibe coded by AI folks. Be wary.
It certainly is, and when it breaks because it can't handle some obscure use case, I won't bat an eye
Install random shit for the promise of .0001ms seek? Jesus Christ guys 🤦♂️.
I am definitely not enabling this, but I thought it was an interesting article. It took Microsoft this long to make a NVME driver.
jokes on you, that measurement became slower with the new driver. but if you look at the other measurements..
why is it that any of your comments are easily downvote worthy? always misleading and/or incorrect.
Issa exaggeration you nerd.
HE SAID SEEK WHEN ITS SEQUENTIAL WRITE PERFORMANCE, GET HIM!!!!!
Just like Linux 20 years ago.
Yeah, but all the Gentoo flags working together would add up to 0.004ms!
You have a problem with my stage1 install?!
Interesting. I presume that over time, MS plans to tweak the driver, increasing safety and security, and then start transitioning known safe devices over. Seems surprisingly responsible.
Only like 10 years late
It's 2026 (basically) and Linux still has major GPU stability issues and doesn't support HDR or vrr over HDMI units using a valve deck image.
Glass houses my man.
for your interest, graphics drivers are made by the graphics vendor. what's more, nvidia made it so that only they can develop drivers for majority of their gpus. I don't see how that's a fault of linux being late.
Doesn't matter. That's the change state, Nvidia being dicks or not.
This is actually kinda cool! Hopefully it eventually becomes a "default" of some sort or at least has an easy toggle in compatible configs.
Here's a reply from an ―allegedly― ex-employee at Msft you can find in the comment section of the article:
There is a lot of confusion on this thread between NVMe Storage Controller drivers and Disk drivers, e.g. "we have always been able to replace NVMe drivers". Previous driver releases, e.g. by Samsung, are for the NVMe Storage Controller, which you don't see in Device Manager unless you view by connection. The inbox driver is "Standard NVM Express Controller" or stornvme.sys. Samsung's driver was secnvme.sys.
The title of this TPU story is misleading; there is no new NVMe (controller) driver, there is a new disk layer driver nvmedisk.sys that is just an optimization of disk.sys that provides marginally better performance for NVMe drives (some SCSI command translations removed; multiple queues supported; presumably latency optimization and cache flush behavior). This is not really an "NVMe driver" because it's not the controller driver. The disk layer driver is not super specific to a particular storage medium; this is just optimization to pair better with stornvme. It's possible that you could force install nvmedisk.sys on HDD and it may even work, albeit unreliably and/or slowly.
Source: I worked at MS for decades. You know that checkbox in Device Manager for drives that says "Turn off Windows write-cache buffer flushing on the device"? That was me.