FlukeThere are differences, alright. You just need to stop testing in HDD-mode which of course didn't suffer much from being dragged through the SCSI-layer.

Here is an example. Lexar NM790 2TB:


And these are mine. Kingston KC3000 2TB:


During AS SSD bench peak CPU usage for my Micron drive dropped from 37% to 29%.

Is it really worth doing though?

The safe mode reg entries would not work on mine red errors, checked and were not added

Are there any benefits enabling this new driver for WD SN850X. The system (with Windows 11 Pro) is mainly used for gaming?

Has anyone seen any Intel Optane benchmarks for this new driver?

warning077Are there any benefits enabling this new driver for WD SN850X. The system (with Windows 11 Pro) is mainly used for gaming?

Imo, no

SoraWD doesn't support dashboard anymore, move on to Sandisc Dashboard.

I did, no difference.

WumygHas anyone seen any Intel Optane benchmarks for this new driver?

No, but HUGE improvements from using Intel's driver vs MS's POS.

FlukeEasy. Also add:

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\SafeBoot\Network\{75416E63-5912-4DFA-AE8F-3EFACCAFFB14}]
@="Storage Disks"

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\SafeBoot\Minimal\{75416E63-5912-4DFA-AE8F-3EFACCAFFB14}]
@="Storage Disks"

Works. Much appreciated.

Also i wanted to see what happens before and after.

1. With nvmedisk.sys set i tried entering Safe Mode with Networking.
Initially it seemed to load, but the Windows logo and spinning circle stayed on screen for bit too long.
It then gave me an BSOD (black screen of death now) with "error inaccessible boot device."

2. Strangely after it went to a restart by itself after BSOD it did boot into safe mode, but this time with disk.sys instead.
So somehow Windows knew and went back to disk.sys:


3. However once i booted back into normal mode it was still using disk.sys again. I thought maybe i needed another restart. I did that but no change. What's strange was that the override values were still present in the registry, but ignored.

4. So i removed the override values, restarted and re-added the override values and restarted again.
Now it's using nvmedisk.sys again and i tested safe mode with the added registry values to confirm it works:


No problem adding these values like user FoulOnWhite had. Perhaps he was not running as admin or something.

(I found O&O RegEditor handy for this sort of thing)

Can someone else check the regedit from techcommunity.microsoft.com/blog/windowsservernewsandbestpractices/announcing-native-nvme-in-windows-server-2025-ushering-in-a-new-era-of-storage-p/4477353?
[ICODE]reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Policies\Microsoft\FeatureManagement\Overrides <span>/</span>v 1176759950 <span>/</span>t REG_DWORD <span>/</span>d 1 <span>/</span>f[/ICODE]

I think in my case it performs ~10% than without it for some tests, but it might very well be run to run variance.

Also reverted back, for the time being. The m.2 cooling fan on the motherboard seem to be a lot less audible using this new driver when playing Halo Infinite (direct storage compatible?), Control (not sure how their "Max" texture quality streaming works?!), and Superposition / various 3d marks, so I'm not sure if that is the lack of BypassIO functionality coming into play? Generally slightly lower scores across the board, although my lowest frames in Superposition did increase (lower CPU overhead?!) but this is fairly anecdotal...

I think I will wait until it is officially part of Windows Update and integrated proper with vetting.. My spidey sense is tingling and thinks Dr. Doom wants to break my PC.

zenlessyankI think I will wait until it is officially part of Windows Update and integrated proper with vetting.. My spidey sense is tingling and thinks Dr. Doom wants to break my PC.

I just figured the "speed" "increases" were not really that much, not enough to risk breaking win 11 any more than it already is.

FoulOnWhiteI just figured the "speed" "increases" were not really that much, not enough to risk breaking win 11 any more than it already is.

Yea I'm thinking the same. W11 is already held together with pins and glue, why poke it?

ncrsUniversal Microsoft drivers have those, supposedly old dates because of the algorithm that governs driver matching and installation. They are providing basic functionality while vendor drivers unlock the full potential. It's not an indicator of driver age per se, but a way to always prefer vendor drivers with newer (real) dates.

From Raymond Chen's blog on the issue:


The default NVMe driver's inefficiency is nothing new. Samsung used to provide their own NVMe driver for improved performance, but stopped doing that since 980 series.

There was barely any benefits, benchmarks were very slightly faster, and NVME is already so fast you dont need an extra 3%. The Microsoft driver benefits from been mature, heavily used, and integrated with the OS. Also the driver has had updates since its original inception.

I personally wont be using this new driver, if its not been automatically used, then its a unsupported configuration. I/O is an area I dont like to take chances with.

FoulOnWhiteWorking on Win 11 Iot enterprise LTSC 24H2
Old driver v new gen 5. up and down old on left. System in Specs

In your case looks like old is better for consumer load, and new is better for server load on performance, did you check power draw and idle temps to see if its idling properly?
Also is CPU/RAM spec same on old/new tests? Q32/T1 is basically a CPU test. For more realism match threads to queue depth. Also perhaps limit queue depth to 4 absolute maximum if consumer use.

chrcolukThere was barely any benefits, benchmarks were very slightly faster, and NVME is already so fast you dont need an extra 3%. The Microsoft driver benefits from been mature, heavily used, and integrated with the OS. Also the driver has had updates since its original inception.

I personally wont be using this new driver, if its not been automatically used, then its a unsupported configuration. I/O is an area I dont like to take chances with.


In your case looks like old is better for consumer load, and new is better for server load on performance, did you check power draw and idle temps to see if its idling properly?

I did not sorry.

Aside from some missing values in some monitoring software i also dont see the downsides as a regular user either.
Yes BypassIO is missing for now, but i dont see degraded game load times or performance and DirectStorage still works.
Safe Mode also works with added registry values.

Also deleting the override values and restarting goes back to the old system.
So it's not like it's permanent. Personally i will take the extra performance anytime even if it breaks some legacy systems.
Im not running server here.

I will continue using it unless i stumble upon a showstopper.

TomorrowAside from some missing values in some monitoring software i also dont see the downsides as a regular user either.
Yes BypassIO is missing for now, but i dont see degraded game load times or performance and DirectStorage still works.
Safe Mode also works with added registry values.

Also deleting the override values and restarting goes back to the old system.
So it's not like it's permanent. Personally i will take the extra performance anytime even if it breaks some legacy systems.
Im not running server here.

I will continue using it unless i stumble upon a showstopper.

What extra performance?

FoulOnWhiteWhat extra performance?

The one i described in my post. In addition to the increased synthetic numbers across three different drives there is repeatable difference in login speed.
Programs also launch faster after login. Disk access with a lot of small files on the disk is also somewhat faster.

I've always moved to new tech reasonably early. I moved to two cores when people were still using one. Same with 64bit vs 32bit.
Same with GPT vs MBR. SSD vs HDD. G-Sync vs VSync. High refresh vs 60Hz etc.

Microsoft will eventually make it the default anyway. The only thing you're doing is postponing the inevitable.
I can understand people who postpone this due to compatibility reasons in a server environment as most of the listed downsides seems to be there.

TomorrowThe one i described in my post. In addition to the increased synthetic numbers across three different drives there is repeatable difference in login speed.
Programs also launch faster after login. Disk access with a lot of small files on the disk is also somewhat faster.

I've always moved to new tech reasonably early. I moved to two cores when people were still using one. Same with 64bit vs 32bit.
Same with GPT vs MBR. SSD vs HDD. G-Sync vs VSync. High refresh vs 60Hz etc.

Microsoft will eventually make it the default anyway. The only thing you're doing is postponing the inevitable.
I can understand people who postpone this due to compatibility reasons in a server environment as most of the listed downsides seems to be there.

I bought a £175 gen 5 M.2 even though nothing is probably going to use the potential 14TB/s read speed, but i just bought it because my board has gen 5 M.2. This driver made near zero difference to that or my still pretty good SN850x drives. If you got a 3% increase fine, enjoy it.

My install works fine, my drives work fine, why do i need to hack windows to use this driver, when it will be released by microsoft anyway officially without no hacking. I will wait.

"I've always moved to new tech reasonably early." Not AM5 or DDR5 though.

FlukeThere are differences, alright. You just need to stop testing in HDD-mode which of course didn't suffer much from being dragged through the SCSI-layer.

Here is an example. Lexar NM790 2TB:


And these are mine. Kingston KC3000 2TB:


During AS SSD bench peak CPU usage for my Micron drive dropped from 37% to 29%.

This is my result on Windows 10 and the old driver... 🤷‍♂️

FoulOnWhiteI bought a £175 gen 5 M.2 even though nothing is probably going to use the potential 14TB/s read speed, but i just bought it because my board has gen 5 M.2. This driver made near zero difference to that or my still pretty good SN850x drives. If you got a 3% increase fine, enjoy it.

I let ChatGPT calculate median percentages based on my results (i know, im lazy).
All drives are Gen 4x4.

For two drives connected trough the chipset the losses were between -0,02% to -0,29% and gains were +3,51% to +3,66%
However for OS drive connected to the CPU the losses were none and gains were +14%.

I had to exclude two outliers from OS drive data. One of them was -2,54% loss and other was +158% gain.

FoulOnWhiteMy install works fine, my drives work fine, why do i need to hack windows to use this driver, when it will be released by microsoft anyway officially without no hacking. I will wait.

Understood. Im not trying to convince somebody to use it. I just saw several posts where people said they were going back so i wanted to add my post about staying on it. Otherwise the impression might be that everyone tried it and rolled back.

FoulOnWhite"I've always moved to new tech reasonably early." Not AM5 or DDR5 though.

That's the exception tho. It's both monetarily and performance wise bad idea to move to a new memory standard in the beginning.
It's better to do it in the middle or even near the end of a memory generation because it brings several benefits:

The standard has far exceeded initial JEDEC specs.
Speeds are higher.
Timings have decreased or at least stayed flat.
Compatibility has been much improved trough numerous patches.
Price has fallen significantly. Unless memory cartels decide to collude or AI companies decides to reserve half of the future production capacity.

My first computer had DDR-400 speeds. I skipped DDR2.

With DDR3 i jumped to DDR3-1866 CL8. DDR3 lasted from 2008-2014 (roughly).
I bought it in 2011 so right in the middle. Initial speeds were DDR3-1066 CL7.
Highest speed kits eventually reached about DDR3-2400 CL10. Lowest latency kits DDR3-1866 CL8.

With DDR4 i jumped to DDR4-3600 CL16. DDR4 lasted from 2014-2022 (roughly).
I bought it in 2019 so more towards the end. Initial speeds were DDR4-2133 CL15.
Highest speed kits eventually reached about DDR4-5333 CL22. Lowest latency kits DDR4-4000 CL14.

With DDR5 i plan on doing the same. Eventually get a low latency kit at the middle or near the end of the generation.
Because for DDR6 it will take years to match and exceed mature DDR5 just like it has been every time in the history.

DDR5 presumably lasts from 2022-2030 (roughly).
I may buy it in 2027 (Zen 6 X3D) so more towards the end. Initial speeds were DDR5-4800 CL40.
Highest speed kits have currently reached about DDR5-9600 CL46. Lowest latency kits DDR5-6000 CL26, but there's more room for improvement.
I have no doubt we will see kits with 10000+ speeds at around CL42-46 and low latency kits at around 6400-7200 CL24-CL26.

Truth be told i have not moved to AM5 and DDR5 also because Zen 5 is not yet sufficiently faster compared to my Zen 3. Even when comparing X3D variants.
Im hoping Zen 6 X3D will be a good step above Zen 5 because Zen 5 was very underwhelming compared to Zen 4.
With rumored move to a bigger 12c/24t CCD, IPC and clock improvements and overclockable (more like unlocked) X3D im aiming for that for my next upgrade. Hopefully by 2027 DDR5 prices have fallen back down (they're already plateauing now) so i can get a low latency kit with which to ride out half the DDR6 generation again.

MakaveliSo I made a restore point and entered the commands and rebooted but doesn't look like the driver change worked on Windows 10.

I think the driver is missing in windows 10 I was able to find it from another source



Just not sure I should continue messing around to try and get it working.

what was the performance difference on W10 ? also notable ? stable ? source ?

TomorrowI let ChatGPT calculate median percentages based on my results (i know, im lazy).
All drives are Gen 4x4.

For two drives connected trough the chipset the losses were between -0,02% to -0,29% and gains were +3,51% to +3,66%
However for OS drive connected to the CPU the losses were none and gains were +14%.

I had to exclude two outliers from OS drive data. One of them was -2,54% loss and other was +158% gain.

Understood. Im not trying to convince somebody to use it. I just saw several posts where people said they were going back so i wanted to add my post about staying on it. Otherwise the impression might be that everyone tried it and rolled back.

That's the exception tho. It's both monetarily and performance wise bad idea to move to a new memory standard in the beginning.
It's better to do it in the middle or even near the end of a memory generation because it brings several benefits:

The standard has far exceeded initial JEDEC specs.
Speeds are higher.
Timings have decreased or at least stayed flat.
Compatibility has been much improved trough numerous patches.
Price has fallen significantly. Unless memory cartels decide to collude or AI companies decides to reserve half of the future production capacity.

My first computer had DDR-400 speeds. I skipped DDR2.

With DDR3 i jumped to DDR3-1866 CL8. DDR3 lasted from 2008-2014 (roughly).
I bought it in 2011 so right in the middle. Initial speeds were DDR3-1066 CL7.
Highest speed kits eventually reached about DDR3-2400 CL10. Lowest latency kits DDR3-1866 CL8.

With DDR4 i jumped to DDR4-3600 CL16. DDR4 lasted from 2014-2022 (roughly).
I bought it in 2019 so more towards the end. Initial speeds were DDR4-2133 CL15.
Highest speed kits eventually reached about DDR4-5333 CL22. Lowest latency kits DDR4-4000 CL14.

With DDR5 i plan on doing the same. Eventually get a low latency kit at the middle or near the end of the generation.
Because for DDR6 it will take years to match and exceed mature DDR5 just like it has been every time in the history.

DDR5 presumably lasts from 2022-2030 (roughly).
I may buy it in 2027 (Zen 6 X3D) so more towards the end. Initial speeds were DDR5-4800 CL40.
Highest speed kits have currently reached about DDR5-9600 CL46. Lowest latency kits DDR5-6000 CL26, but there's more room for improvement.
I have no doubt we will see kits with 10000+ speeds at around CL42-46 and low latency kits at around 6400-7200 CL24-CL26.

Truth be told i have not moved to AM5 and DDR5 also because Zen 5 is not yet sufficiently faster compared to my Zen 3. Even when comparing X3D variants.
Im hoping Zen 6 X3D will be a good step above Zen 5 because Zen 5 was very underwhelming compared to Zen 4.
With rumored move to a bigger 12c/24t CCD, IPC and clock improvements and overclockable (more like unlocked) X3D im aiming for that for my next upgrade. Hopefully by 2027 DDR5 prices have fallen back down (they're already plateauing now) so i can get a low latency kit with which to ride out half the DDR6 generation again.

Isn't C28/30 considered low latency for DDR5, it seems from reviews i have read, systems seem to perform at the sweet spot with 6000C30 which is what mine is.

I switched because at the time the 7800X3D was/is the best CPU for gaming which is what i mostly do, and it is better than my previous intel setup. (I have no bias eother way, i go where the best performance is) My board is at least ok for zen 6 i believe so i can drop in some zen 6 X3D CPU and still be good for a while too.

Imo a 7800X3D is significantly faster than a zen 3 5800X3D setup.

No matter the sticker on the SSD; there are only 3 or 4 SSD controller chip makers.
That means that:
Any driver, written for one of those ctlrs, will work on ALL drives that use that same ctlr.
"BUT THEY DON'T!" I hear you say. 😄

The actual .sys driver works.
The plain text, easily editable .inf file is where the driver writer specifies: "Only for OUR drives!"
So if the .inf file is edited to add your drive's name to the: "Only for" list; it works.

Yes, but the whole driver package (.zip or whatever) is signed by it's hash value.
You have to sign the driver to get it to work.
Or:
You can just read the 'How to's' and download a mod+signed driver from the Win-RAid Forum. (Fernando)


Pre Linux, I was running the Samsung driver on a 500GB Corsair MP600:

Stock MS Driver:

Samsung Driver:


Seems on par with the discussed driver, WITHOUT all the "not working" drawbacks..?


MS driver vs Intel Driver on 58GB Optane 800p (on AMD System..!)

Dwardenwhat was the performance difference on W10 ? also notable ? stable ? source ?

It doesn't work on Win 10 as the driver file is missing.