"The AI chip ecosystem is even more diverse, with Windows 11 PCs supporting AI chips, CPUs and GPUs from Intel, AMD and NVIDIA.
That’s where Windows ML 2.0 steps in. The Windows ML runtime handles the heavy lifting of identifying the hardware and automating hardware support for AI models, while also extracting the maximum performance from AI chips.
Windows ML 2.0 also figures out dependencies, procurement and update management, and includes them in installers. That is typically a lot of manual labor in fragmented environments."
I think DXO works with ML and not with ML 2.0. That would also explain why there are problems with driver versions in combination with different graphics cards. The interaction of all combinations must be managed and controlled by DXO—which is complex, as the current situation shows.
Public Service Announcement: I would be reluctant to use PL9 on Win11 PCs with Nvidia graphics cards.
I installed PL 9.02 on Win11. I felt a bit misled by the message “always make sure to install the latest graphics driver from your manufacturer” (RTX 3090 here, 24GB GPU RAM).
Contrary to the release notes, nothing is fixed about the fundamental GPU incompatibilities. The internal error messages always pop up when using scene selections (e.g. ‘background’). Opening an image that was saved with these kind of selections will prevent it from being edited at all. The issue occurs irrespective of image size (e.g. error appears even with my old Canon 5D .CR2 files from 2006).
Using the CPU (in my case a 32 core Threadripper) is way too slow. Even with 24MP images. PL 9 is significantly not workable in its current state (on Win11/Nvidia), if you are buying it for the new AI features.
2 Likes
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
325
I have made some more test using the main two methods of using AI-masks:
The predefined AI-preset masking method AND The freehand-hoovering-select-method and found some interesting results.
It is the use of the predefined AI-presets that is causing all the problems with “Internal Errors” we have seen so far with AI-masking. These problems seems simply not be there or occur when using the other method and in the link below I have done export tests with two RAW-pictures developed with these two methods and edited them in simalar ways. The results are very clear: The export with the premade-AI-preset edited picture took FIVE TIMES LONGER TIME than with the other method.
Stenis, this is consistent with my experience as well. The only issues I had were when attempting to use the keyword-based AI presets vs hover and click. When I finally re-arranged drivers and settings and could export (but not view or edit) the photos with the preset-based masks, the export took much longer than when I used the hover select. I didn’t time it like you did but I can confirm I noticed a similar difference.
2 Likes
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
327
Before DXO has adapted their code to the present new Nvidia-drivers, using the premade AI-presets in a NO GO-choice for very many users. For some it will crasch instantly, for others it will be very slow and/or will cause problems when saving or printing.
Use the other method instead - it works for many, is very stable and much faster. It works even with the older FTX-cards and seem to work totally flawlessly too.
Care to elaborate? I appreciate all the effort the PL9 users have put in to figure out this problem. But is it a problem for everybody, or only for the ones that write about in this post? I am getting the AMD 9950X3D and a 5090 pretty soon, just to be able to run the most advanced features in PL9 and On1 without problems or slow processing. Are there any other DXO / Photo-forums that address these issues? From DXO support: “This problem is caused by the latest NVIDIA drivers in the 580.xx series”. The DXO company seems not to really care but instead blames it on Nvidia, that is is worrisome.
I just sendt DXO-support this message: "When is DXO going to fix these problems that you are aware of instead of blaming the largest GPU producing company, Nvidia, on the earth? If DXO does not realize that Nvidia is NOT to blame, DXO IS to blame, you WILL loose most of your PL9 customers, including me.
How do you know that Nvidia is not the problem? Another user here has got the message from their support that it was a known issue with their graphics driver.
Somehow, in the last 8 days, MS has now qualified their part of the software and shifted the onus on the users. No formal release notice or notes on changes. So, when is a software “release” not “beta”?
What MS says is: we provide environment, some basic code, and a “glue”, but we take no responsibilty for what you stitch. Maybe the text is there to be safe with the lawyers, like the following from Nikon D4 User Manual:
" Notice for Customers in the State of California WARNING: Handling the cord on this product may expose you to lead, a chemical known to the State of California to cause birth defects or other reproductive harm. Wash hands after handling.
"
I’m using Intel Arc gpu (580, 12 gb VRam) and facing same problems. Nvidia is not the problem.
2 Likes
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
338
Maybe DXO already would have fixed these problems even before the release if they had used the 580.xx drivers when developing version 9. One of the main ideas with a common driver interface like Windows is not like before having to develop dedicated drivers for every application - because that is really costly. That means though in short that the application developers have to relate and adapt to the drivers provided by the hardware-manufacturers and not the other way around because if the later will reign the whole idea with a common driver interface like Windows would be compromised. I don’t think the software industry really would like to see that swamp to emerge because that would also increase the costs a lot for the software developers.
Doesn’t DXO already have enough of that developing a lot of profiles for cameras and lenses and testing many many thousands of these combos just because the industry refuses to standardize?
I remember the days of the nineties when Wordperfect had to developed and maintain printer drivers for the whole printer market since it had 60-70% of the word-processing market.
Of that follows that it is DXO that has the main responsibility to fix these problems we are having with Photolab 9 and not Nvidia.
I agree. But as I have gained at least some insight into the matter, I think, Nvidia should also be interested in finding out what the problem is, for their own sake. Being blamed for a software that does not work on their hardware is not good for business no matter who is at fault.
It’s interesting that 572.83 is generally recommended by game developers (and @Stenis here), at least for GeForce RTX 50 series, while DaVinci support says it has memory leaks and was blacklisted recently:
Starting in Resolve 20.1.1, a check was put in
for this specific driver version,
due to discovered memory leak issues with Resolve,
so no Resolve version of 20.1.1 or later
will let you use the 572.83 driver.
BTW, 566.36 is “generally” believed to be the only stable version for older NVIDIA cards, e.g. GeForce RTX 40xx, also recommended by game developers, while some people had to revert to even older versions. DaVinci Resolve probably uses GPU functionalities used both by games and PhotoLab, but maybe PL9 uses also something still rarely used (?). It’s all getting too complicated…
