Current Versions, stable Graphics Drivers for Windows 10/11, 26 Feb 2026

NVIDIA GeForce Graphics Drivers 595.59 WHQL for Turing and newer GPUs (26 Feb 2026)

AMD Radeon Graphics Drivers 26.2.2 WHQL (26 Feb 2026)

Intel Graphics Drivers 101.8509 WHQL (for Arc/Xe/Core Ultra) (16 Feb 2026)

Intel has dropped support for and frozen video drivers for UHD Graphics (5th-10th CPU generation) and UHD/Iris Graphics (11th-14th generation). No further functional improvements or bug-fixes will be released although there may be some intermittent security fixes released if Intel judges the problems to be “severe enough”.

UHD / IGP Drivers for 6-10 gen CPU; frozen at gfx_win_101.2140
UHD / IGP Drivers for 11-14 gen CPU; frozen at gfx_win_101.7084

In Dec 2025, Nvidia froze support for any GPU older than Turing (meaning Pascal and older, GeForce 1050 Ti and older). So if your GPU is based on Pascal, Maxwell or Volta, you’re pretty much frozen at level 582.28 which is a security update based on 581.80. NVidia release notes and their on-line driver search tool have been updated and are the only authority worth referencing for older cards / GPUs.

In the near future, Nvidia will be dropping the game / studio delineation of driver releases and thus put to an end the needless arguments and misunderstandings over what those terms mean.

Note that both Intel and AMD publish separate driver packages for motherboard/chipset/cpu, networking, and native disk/storage controllers.

Note that Intel publishes separate driver packages for on-board NPU functionality on Core Ultra processors (enabling on Win11 only).

Note that Intel will often rapid-fire beta graphics drivers to provide support for specific gaming environments on Arc/Xe graphics.

1 Like

@Sparky2006 Thank you for the update I downloaded the drivers last night to my 5600G(32GB)-3060(12GB) machine and got the following results with the St Pancras image late last night (actually earlier today).

!!UPDATE:- The drivers have been pulled!!

The image is extremely large and the RTX 3060 is a member of the RTX x060 group of GPUs which DxO have indicated is having problems with AI masks!

All exports were done with a single export worker to minimize the risk of complications with the export process, well that was my “theory”!

PL9 is now coming up to its 6 month birthday and I had high hopes after I got this with my “Renamed” test group, i.e. the image is copied and the name changed to reflect the AI mask being used, which appeared to work successfully.

Switching to a duplicate of the same test group but with XD2s Noise reduction selected I had a less successful outcome

and repeating the XD2s tests with a test group composed of a single image + 6 VCs I got this result

So I restarted testing today having cleared the database and got the following for the NO NR tests with VCs versus separate images and got the following which was not very encouraging but at least the same edits failed on each test (this time)!

As for the XD2s tests they were even worse

I will attempt to repeat the test on the 5900X(32GB)-5060Ti(16GB) but that normally performs even worse in these tests!!

Still using the latest Nvidia Studio driver 591.74 – works reliable so far.

@Wolfgang None work on either of my machines typically the 5060Ti has more failures that the 3060 and neither has none except that rare example which I reported in the post above!

The intriguing (or should that be infuriating) part is the lack of pattern in the errors, i.e. where in the cycle they occur. So this is the first and second test on the 5060Ti(16GB) with 1 worker, empty database and export worker terminated between each export.

The drivers on that machine are 591.74

and the NO NR Test results look like this

The name has changed because I got concerned about getting confused with the snapshots from System 3(5600G-3060) and System 1(5900X-5060Ti) so DSCF0668 has becomes DSCF0688-1 (i.e. residing on System 1).

@Wolfgang Not very useful when every image with a mask fails!

With XD2s things are identical except no the reported cause of the error

But because the export worker is left running after the first error basically nothing is going to work after the first error in this case.

Bryan, all I can say is that PL9.5 is running, not the fastest, but it works.
( And examples without masks run on PL8.x faster than on PL9.5 )

.

I see some differences to your screenshot, but I’m not technically savvy enough to interpret them correctly.

I am hoping that GPU-Z or HWInfo will be updated to account for WinML versions and NPU/TOPS.

WinML has become as important as it’s predecessors of CUDA, OpenCL, OpenGL, and Vulkan. That it has to be supported across multiple GPUs and now multiple NPU generations has proven a challenge for Nvidia although AMDs support seems to work well.

In their buried developer forums, Nvidia is still struggling with their implementation and support for WinML including vram management and garbage collection.

1 Like

Same version here, but Win 25H2 brought some new problems – driver resets with 2-3 sec black screen while using Web browser or Photos app. Looks like a well known problem.
I still have problems (using RTX 4070 on Win 26200.7840) with AI masks sometimes. They look like some timing issues with memory allocation involved – sometimes GPU native memory usage goes up to 11.6GB (out of 12GB). If ‘Shared GPU Memory’ usage goes beyond 5GB, I usually have to restart PL9.5 to get standard export times. The “Shared GPU memory” is managed by Windows, so it looks like it’s a really complex problem involving Microsoft/NVIDIA (and DxO). I’ve predicted that it will be solved by spring 2026 but my hopes are expiring. After the fix provided by NVIDIA few months ago and some undocumented Windows changes, I’m able to get my work with AI masks done, but it’s not comfortable. To summarize, with 591.74/25H2 there are still ocassional black screens and GPU memory leaks related to AI masks, but for non AI masks PL9.5 works stably.

It’s high time for DxO to provide some feedback from MS and NVIDIA, unless they are under NDA.

Spring is not over yet!

also


see …

The NVIDIA Studio and Game Ready drivers 595.59 have been removed:

February 26th, 11am PT Update: We have discovered a bug 
in the Game Ready and Studio 595.59 WHQL drivers 
and have removed the downloads temporarily while our team investigates. 
For users that have already installed this driver 
and are experiencing issues with fan control, 
please roll back to 591.86 WHQL.

See Resident Evil™ Requiem GeForce Game Ready Driver Released

There’s a new version of game ready driver, 595.71 released on Mar 2, but it already has a hotfix 595.76.

@Wlodek I used those (595.76) on my 3060(12GB) and still have errors with the St Pancras image.


Interestingly or not) the first test after the installation yielded no errors

but every tests since then has faulted on the ‘Vehicles’ AI preset.

ALL tests are NO NR, single export copy. All have been after a PL5.4.0 restart and some after the removal of the database as well, i.e.

Install
Test successful
Restart PL
Test failure
Restart PL and remove database
Test failure
etc. etc.

@BHAYT

Your GPU-Z screenshot shows 559.59
Did you mix up the screenshot or is it the faulty driver (see @Wlodeks’ comment) ?

I’m still on “Studio” 591.74

@Wolfgang I was obviously sleeping on the job when I took the snapshot.

This is the actual end of loading the new drivers


and the date/timestamp predates the snapshot I used.

The reason for the problem is that GPU-Z was running before and throughout the driver update but retains the original GPU details until forced to refresh and I neglected to do the refresh before taking the snapshot - sorry.

@Wolfgang Thank you for your diligence I will try not to make that mistake again!

In the meantime I have had a request for the original image which can be found here, I believe, https://www.dpreview.com/sample-galleries/4314410403/fujifilm-gfx-100-ii-pre-production-sample-gallery/2913371124 and was discovered by @Wolfgang.

It continues to cause problems with both of my machines with a 3060(12GB) and a 5060Ti(16GB) GPU (arguably both of the x060 models, although whether the 5060Ti behaves the same as the 5060 I don’t know).

The re-ordered (from @Wolfgang’s original) DOP is
DSCF0668.RAF.dop (85.4 KB)

The DOPs for testing individual files are
00-DSCF0668.RAF.dop (9.8 KB)
01-Sky Preset.RAF.dop (11.2 KB)
02-Vehicle Preset.RAF.dop (11.2 KB)
03-People Preset.RAF.dop (11.0 KB)
04-Building Selection.RAF.dop (14.2 KB)
05-1+2+4 .RAF.dop (16.3 KB)
06-1+2+3+4.RAF.dop (16.1 KB)

The image will need to be copied multiple times and given the following names

00-DSCF0668.RAF
00-DSCF0668.RAF.dop
00-DSCF0668.xmp
01-Sky Preset.RAF
01-Sky Preset.RAF.dop
01-Sky Preset.xmp
02-Vehicle Preset.RAF
02-Vehicle Preset.RAF.dop
02-Vehicle Preset.xmp
03-People Preset.RAF
03-People Preset.RAF.dop
03-People Preset.xmp
04-Building Selection.RAF
04-Building Selection.RAF.dop
04-Building Selection.xmp
05-1+2+4 .RAF
05-1+2+4 .RAF.dop
05-1+2+4 .xmp
06-1+2+3+4.RAF
06-1+2+3+4.RAF.dop
06-1+2+3+4.xmp

Unless you have some old games from 2015 or older that need it, you can stop installing PhysX.

@Sparky2006 I have had trouble when trying to go back to an older set of drivers with a report that I am trying to install an earlier version of PhysX, and the installation just terminates.

However, why is installing PhysX a no-no and how do I prevent it being installed?

It’s part of the driver. So why trying to exclude it??

@Wolfgang I wasn’t except that @Sparky2006 had stated

and I was responding to that message but PhysX has prevented some of my attempts to install earlier drivers and I didn’t know that I could stop the installation and was wondering why I would want to!!?

Well, do what you think is right. – I’m not technically savvy enough to know the details, but stick with the Studio driver as long as it works … which it does.

.

I wonder if the problem with

Note for NVIDIA RTX x60 models (2060 / 3060 / 4060 / 5060)
These models are affected by a known WinML issue. Until NVIDIA releases a fix, you can:

  • Use CPU processing for these operations, or
  • Avoid AI Mask keywords temporarily.
    We are closely monitoring the vendor’s update.

has already been addressed … perhaps this is your problem too.