I have an image that works on my 5060Ti(16GB) and used to fail on my 3060(12GB) so I just tried it with PL9.1.0 and the latest NVidia drivers 581.42 Games Ready and
it still fails!!
Try to downgrade your drivers till v577.00 or lower. There are some issues with recent Nvdia drivers unfortunatly:(
It still says this on the release notesâŚlooks the âfixâ is still awaitedâŚ
Known Limitations
⢠AI Masks/AI Keywords errors with NVIDIA GPUs: With NVIDIA drivers in the 58x.y series (e.g. 580.97, 581.08),
AI Masks and AI Keywords may stop working after a few seconds, requiring a restart of PhotoLab. Please use
57x.y drivers (e.g. 573.24, 577.00) as a workaround. A fix will be provided as soon as possible.
@Longwaytop Welcome to the forum but obviously you are new to the forum since I have written about this extensively for PL9.0.1 and PL9.0.2 and while my 5900X with RTX5060Ti(16GB) manages one particular image with no problem but my 5600G with 3060(12GB) does not.
So as soon at the new PL9.1.0 release came out I located the very latest NVidia drivers in the hope that something (anything) might have changed for the better but it hasnât so after a little more testing I will return the 3060(12GB) back to the earlier version where it works.
But thanks for the heads up anyway and donât be put off commenting by an âold lagâ like me.
@BoxBrownie Thank you for the heads up, I thought I would test it anyway because I did check the release notes but didnât read to the end and its is so important they wrote it twice
1 Like
The one comparison in âfeedback to the customerâ I can make is the way Affinity address their updates with a comprehensive list of additions, fixes & outstanding problems to be addressed.
So, when DxO say this
Bug fixes
â˘
Minor bug fixes and improvements
It is neither informative or confidence building, though for the use I have made of PL since IIRC v4 I do like it for various of the features including the newest AI masking. Of which I am still getting to grips with but still cannot see how the âMâ key works in showing/hiding the active mask.
I was advised on another thread that it works only once I have selected a/the mask in the Mask Palette but pressing âMâ does nothing. Câest la vie the learning is ever onwards 
3 Likes
They have refused to tell any one, at times its appears even support has no idea and have asked me if a bug I reported has been fixed
They cant even be bothered to number updates after doing so for a short time
When updating software DxO says that one should ALWAYS use latest drivers. This is DxOâs advice!
@leinmi It still is, in the release notes, except for the caveat at the end when it is suggesting an alternative strategy, i.e. to use older drivers.
With the one image that I have that is causing, issues my 5060TI(16GB) is fine on what was the latest drivers until the one I tested today (which I havenât tested with the 5060Ti yet) was released.
But my 3060(12GB) does not work with the latest drivers or the version just before, at all but is fine with an earlier version
So the rule is to use the latest drivers unless they donât work and then either use the alternative set recommended by DxO or hunt out an earlier set that will work with your GPU.
1 Like
That is standard boilerplate text that DXO always uses, however regardless of that being in the release notes, ignore it for the time being and instead use an earlier studio driver like 572.83 which many of us are using. successfully, until the driver issue is worked out between DxO and Nvidia.
According to a user here who had a conversation with Nvidia support, they acknowledged to him / her that they are in the process of trying to revolve the driver issue which seems to suggest that it is not just a DxO coding problem.
Mark
Version 9.1 is definitely an improvement and it seems now that it is actually working as expected for me despite an old 3060 Ti card at least as long as I use the Nvidia 572.83-driver instead of the newer ones.
To the tread owner: Maybe you shall try the Studio-driver instead despite it might not be what you want to hear if you are a gamer too :-). Below you can read my experiences. This is just an initial test of some pictures but despite that I think version 9.1 is promising but the results needs to be verified or falsified with other pictures too.
New service release Photolab 9.1 - Issues? Solved any problems? - DxO PhotoLab - DxO Forum
What C1 has done is to invite the users to a lot of webinars:
and they have a lot of good educational videos too:
But I think it would have been really bad if something like the problems we have had with release 9 had happened with Capture One.
DXO has just chosen a different way . They still has an open forum but DXO doesn´t give it much love so here we just have to help ourselves and that doesn´t cost DXO all that much more than waht Caoture One spends on their users when it comes to a user dialog.
1 Like
@Stenis I am the thread owner and the Studio versus Games Nvidia driver discussions have been going on forever. I have run one machine on one and my other on the other and never found one more of less stable than the other.
When reviewing suitable drivers for the 5060TI(16GB) I found that there were more Games Ready drivers in the list of potential drivers than there were Studio Drivers, i.e. between a release there might be up to three additional Games Ready drivers before another Studio driver was released.
The reason for having to look at drivers was because NVidia would not install the recommended earlier driver on the 5060Ti.
So I found the earliest Studio driver and installed that and it worked and then tried the latest Studio driver and that also worked, i.e. the 5060Ti(16GB) does not suffer the same problem as the 3060(12GB) card does on the later drivers!?
Since I had the 3060(12GB) running today when both PL9.1.0 and a new driver from NVidia, a Games Ready driver as it happens, were released at roughly the same time I put 2 and 2 together and made 4, but it the turned out that the answer on the 3060(12GB) was 3 or 2.5 or âŚ
PL9.1.0 does not fix my problem on the 3060, the âSkyâ AI mask breaks it immediately and I donât for one minute believe it is a memory issue.
If PL9.1.0 works for you then that is good, although from your earlier posts I believe that PL9.0.2 was also working for you.
With respect to what any other Software companies do I really donât care, my product of choice since OpticsPro has been DxO. Sadly they released PL9 prematurely and failed to undertake proper Release Candidate testing, or so it appears, and brought the release date forward ,which has helped no-one.
Not arranging to test the proposed Release on the latest drivers and on a variety of kit is a ridiculous state of affairs. PL9 is a release with a lot of new stuff so it needs a lot of structured testing and I am afraid I have seen precious little of that from DxO.
Guiding users to undertake such tests needs the developers to come down from their ivory towers and out from behind their castle walls and mix with us âpeasantsâ and organise, digging the ditch and erecting the barricades, i.e. utilising the talents and the kit of the users to thoroughly test product elements on as wide a set of kit as they feel necessary.
3 Likes
An Update from three runs on the 5900X with 5060Ti(16GB) with the image that cause my 3060(12GB) such grief.
It didnât get a snapshot of the old drivers I was using, amateurish sorry.
UPDATE: But I believe they were 576.02 because the NVidia installer would not install the 576.23 drivers on the 5060Ti GPU.
Then on Studio Drivers
Then on the latest Games Drivers
All memory usage figures are high and the timings get progressively worse!?
1 Like
FWIW
I am using Studio driver version 576.02 and all is (apparently) well with my 3060 (12GB) GPU
@BoxBrownie After my export runs on the 5900X with the 5060Ti(16GB) which I reported last night I transferred the directory of images used to the 5600G with 3060(12GB) and loaded 576.02 which I believe was the set of drivers that were installed on the 5060Ti for the first set of tests in my post above.
On both system the images are on the same make of NVME (but one machine has an MSI motherboard and the other an ASUS Motherboard) and the exports are to the same NVME.
The figures for the test on the 5600G(3060) are in the following snapshot
and I then returned to the 5900X(5060Ti) and installed the same driver and repeated the test and we have
In both cases no looking inside the edits etc. just ensure that the correct Noise Reduction was being used and straight into the export of 20 images. In both cases on opening the test directory the wait for the first image to be rendered seemed long!?
The 5900X took 14293 MB and 1 minute and 13 seconds (73 seconds).
The 5600G took 12093 MB and 2 minutes and 20 seconds (140 seconds).
So the PL910 appears to run in âonlyâ 12093MB on one machine but uses 14293 on the other and the settings on both were
PS:- if anyone is interested this is the image and the DOP (with very crudely applied AI masks)
P1012188.RW2 (23.1 MB)
P1012188.RW2.dop (261.4 KB)
P1012188.xmp (1008 Bytes)
Any performance improvement noticed in PL9.1 (in comparison to PL9.02) for WIN 11 installations?
@adi99 I should have known someone would ask that question but my tests are all Win 10 so it would need someone with Win11 to use my image and DOP from above and test on PL9.0.2 first and then upgrade to PL9.1.0 and then there is the choice of drivers. But that is a 9.1 DOP and may or may not be accepted by PL9.0.2. This is a 9.0.2 DOP from an earlier test group which should be similar or the same.
P1012188.RW2.dop (224.3 KB)
I will save the DOP and rerun my tests on PL9.1.0 to get some performance figures when I have more time.
576.02 currently seems the safest for 3060(12GB), it would be interesting to see the image and DOP run on a 3060(8GB) or any other 8GB GPU.
The fact that my 5060TI(16GB) takes more memory to run the same test group as my 3060(12GB) seems to hint at the fact that currently DxPL is taking whatever memory is available on the GPU, which I has suggested might be the case in other posts, i.e. it may well be able to run with less memory but it will take whatever is available?
PS:- This is a straight from âSleepâ test which took approximately the same time as the previous run but using the PL9.0.2 DOP and the results seem to match
but the GPU memory usage is only 7482MB!
Update:- I just turned on my 5600G with the 3060(12GB) and opened PL9.1.0 and it went into the last export test that I had done and immediately fell over and it definitely wasnât memory problems, at least not the quantity available!
So ignore my advice about using driver version 576.02, although I will leave it on the 5900X with 5060Ti(16GB) for now but will return the 3060(12GB) back to 573.24.
I even went back to the âgame readyâ version 528.24 to get PL9.1 working normally and quickly again 
1 Like
I have a even stranger problem. During image export from NEF to JPG I always have âcorrection failed on execution phaseâ and that for all NEF files where I applied any of AI masks AND I modify some image setting on this AI mask. Non AI masks work OK. It seems that problems start after PL9.1 manages to use all of available 8 GB of GPE memory (RTX 4070 notebook, studio driver 577.00). It is always happening when GPU ram is at 100 %. It appears PL9.1 doesnât release or manage GPU memory properly. Any similar experiences?
GPU memory is mostly managed by the driver.
PL can only try to make workarounds for that.