Off-topic:
With quite complicated AI masks it’s up to 3 seconds here (with DeepPRIME previews enabled, using DP3 or XD3). With no Local Adjustments it’s almost instant. I don’t think I would tolerate a 30 seconds lag…
BTW, C1 has very simple denoising, while DeepPRIME can hit the performance severly on low-end hardware.
Don’t think anyone will argue with you there. However, simple observation of how DxO operate strongly suggests that financial factors mean they are driven by the constant need for new users, and that means a constant need for new features, to grab headlines. Optimisation requires time and effort to revisit existing stuff. While such effort would be noticed and welcomed by loyal users, such effort is never going to grab headlines, since it’s halfway to admitting stuff was only ever half-baked.
I’d love to be proven wrong on my above assessment.
(I tried to DM but it wouldn’t send, so…)
It sounds like you have very minimal slowdown which sounds like a wonderful place to be. I don’t mind saying my PC isn’t the newest - but it can cope with pretty much anything I throw at it. PL is the worst performing app I try to use.
Can I ask what kind of system you have?
You’re right about C1 having simple de-noising (and from my experiments it can’t export anything nearly as sharp as C1 can). Honestly those features are what keep me “loyal” to PL. Finished shots in PL can be so much better than anything else I’ve tried.
If I could sort out the performance issues, most of my major complaints would vanish and I could happily beg for AI object removal etc. with everyone else 
Win11pro 26200.8037, i7-14700KF (sadly without iGPU), 32GB RAM, GeForce RTX-4700/595.79, 4K 27" monitor, 2TB+4TB nvme disks, mostly 45mpx Z8 raws.
It’s only 2.5 years old. I usually rotate my desktops after 7 years or so. My previous desktop “survived” for 12 years though, but that was HP Z220 beast with 64GB RAM, Xeon CPU, Win7. With this pace of “AI” growth, my current desktop will not serve that long, although it looked a slight overkill for me when buying…
Got it thanks, yes your system is definitely an all-around upgrade from mine - your CPU is around 9 years newer and has 20 Cores, 28 Threads to my 4 Cores, 8 Threads (it is in fact a Xeon CPU that is still chugging like a beast, considering). I imagine you’re running faster RAM too (mine is similarly old DDR3). My images are only ~20mpx R6 RAWs so I can only image how much slower my system would be in editing 45mpx ones.
I don’t relish the idea of having to upgrade often to keep up with the latest software trend (it’s just not economically viable as my other ‘hobby’ is a car!) But perhaps I should upgrade soon.
Yes. Microsoft going to more streamline the AI as part of the OS, and they was not too happy about nVidia quality in consumer GPU side (nVidia execution provider for WinML).
DxO reasonably relies on some high-level interfaces
Yes, you pinpoint (or bullseye? what is the correct English term?) the thing. All this part more-or-less ‘transparent’ for DxO. If someone (nVidia, Intel, etc.) has a provider for WinML, its just expected to works, doesn’t matter its GPU or NPU or whatever accelerator the future holds.
I would like them to do what Capture One has done years ago with their lovely sets of Smart Brushes that automatically adds a layer when we select one of them and start using it. It is a very natural way of doing it that works especially well with a pen and pad.
I also agree on everything @Wlodek wrote above. That was spot on. I also love the face editing tools in C1. Hope for something similar in Photolab.. Otherwise I love it as it is now.
This all gets oh-so-much-worse when masks are introduced. At times I’ve found the “generating preview” task doesn’t actually complete at all, or takes more than half a minute to complete. All that when changing something relatively minor in a mask.
Send a raw+dop via something, and i try with my ancient 4GB GPU 
Note: please not with Pre-defined AI mask (i can’t handle it with 4GB GPU). I’m not question the lag and delay, just curious → and i let you know how its behave in here.
First wish : when opening a raw, update the '“is processed” flag.
Why ?
I work in project mode ; if I process a raw and then move or rename it, PL forgets that it has been processed.
All other indicators of dop file is read and updated in the database.
Second wish : process panoramas.
I would like that PL includes panorama stiching : buy the source code of Autopano !
Third wish : don’t limit the size of pictures that PL can process at that low level.
When I process a pano, I have an error message that limit size is 30000 px => in fact it can be much lower, specially with reshape tool.
The ‘is processed’ flag is related to exports rather than just image reading or edits.
During export the status changes as follows:
‘Awaiting processing’ → ‘Processing’ → ‘Processed’ or ‘Processing error occured’.
For me, it’s elliptical control points before anything else.
It seems you haven’t understood the reason of my wish….
Support for monochrome sensors.
Polishing, indeed, particularly for the UI. On an Apple Studio Display set to Photography mode at native resolution the non-active panel options are very difficult to see where a dimmed item appears to be non-actionable as is typical of, oh, every other software UI but with PhotoLab an item is ‘live’ but not being used the result being overlooked opportunities and a lot of extra clicking around. Options for larger text and varying the gray/black shades in the menus also desired.
I would really like to get a mobile version of Photolab (like LR has since years).
When travelling, I usually leave my computer / laptop at home and just take my tablet computer with me.
Therefore a tablet verion of Photolab would be very useful. I could accept both, ipad or Android tablet.
Both platforms offer high performance tablets (e.g. Apple iPad Pro or Samsung Galaxy S series) which should have enough computing power to run PL.
Really? Are there any tablets that have enough GPU power to allow the use of the AI masking features and / or enable exports to complete in seconds rather than (many) minutes?
The iPad Pro has an M5 processor with 10-core GPU and the iPad Air has M4 with 9-Core GPU. People on these forums have been using PL’s masking on MacBook Pros with similar specs without issue, and it’s the exact same chips in these iPads, so there should be no issues there.
Ever seen the spec of an iPadPro?