Regarding crash:
Even the DxO release not not say about: Pre-Defined AI mask, DP XD2s.
May worth to try to see what’s happen when use:
NR set to DP3
Deepprime rendering: off (uncheck)
Export only 1 (one) photo at the time (set paralel export to 1, and not default 2).
Close any other app (like web browser).
Not use Pre-defined AI (like ‘Sky’) mask - but like manual ‘selection AI mask’
DP XD2s, Deeprime rendering, paralel Export (like 2 or more) and ‘Keyword AI mask’ each use a bit more GPU VRAM over the DP3, etc. So, ‘more quality’ use more GPU VRAM. Image something like: DP3 use 1.5GB, DP XD2s use 2.5GB VRAM (not exact values, but to get an idea). I try to calculate this more exactly (previously, with 9,0, 9.01, 9.1 version) and i think for 6GB GPU VRAM looks Okay for DP3, etc. If everything is ‘up’, may even 8GB VRAM cant be enough.
Regarding GPU VRAM memory ‘not released’ - i think its normal / okay. Export process can be killed (DxO.PhotoLab.ProcessingCore.exe) after export (its safe, its restart itself) - and you see GPU VRAM usage drop a bit. But in the end, when you export again, its use up GPU VRAM, so overall its doesn’t matter too much if any.
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
22
I have problems too:
I tried to apply a “Person”-premade mask from the menu and got the answer (after AI tried to do the job that it could not fins any person in this motif and then it crashed as it it rarely did with the prior version.
PL9.2 AI masking seems a bit better for me than 9.1 but, as others seem to be finding, it still consumes and doesn’t release GPU VRAM.
I did a simple test on my Windows 11 Intel i7-14700 CPU with 12GB RAM, NVidia GeForce RTX 4060 Ti with 8Gb VRAM
Took a RAW image of a flower that contained simple corrections - Colour Balance, Smart Lightining, Contrast - and created a VC.
Exported the Master as a 100% quality JPEG - export time 5s
Applied an AI Flower subject Mask to the VC (and dramatically reduced the Local Adjustment exposure to make the affected area very visible).
Exported the VC with the same export settings as for the Master - export time 20s
Selected the Master (i.e. the one with no AI mask) again and exported as before - export time 25s
Windows Task Manager shows 7.7/8Gb consumed
Close and relaunched PL9.
Exported the Master again - export time 5s (and Task Manager shows 0.5Gb/8Gb consumed.
My conclusion - at best, once you bring an AI mask into play PL9 export (and editing) performance will decrease significantly until you close and reopen PL9.
@andras.csore The important aspect is not the number in the batch, although in this case the failures have moved from the first two to later in the batch but it appears to be caused by swapping export types, typically between DP3 and XD2s, but not necessarily in that order, from one export run to the next.
I believe you have picked a 5070? The 5000 series appear to be immune to the problems that seem to occur with other Nvidia cards, even the 4000 series appear to suffer problems and then you get this
I will switch computers and see how the 5060Ti(16GB) behaves. This new release might start breaking even that GPU.
Looking at the edits for those 20 “images” there is something very wrong. I can’t remember the old test settings but these are a long way from what I expected
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
31
This “upgrade” is not at all for me!
It has really fucked up my computer and now even the freehand AI is sluggish and crashes, so I will go back to 9.1.
It used to be five times faster.
With this version I can´t work at all as it looks
I just created a new directory so that I can leave the original one as it is and start afresh
What does seem to have got lost on a number of images I have been transferring between machines is the LA edits I put in so the AI would have some work to do.
They are “obvious” in the thumbnails but they have gone in the move of the PL91 images to PL92 on both machines.
@andras.csore I really did know that and I use GPU.Z to monitor VRAM usage, typically the maximum, used but its graph can be set to show what is currently being.
The repeat test on the 5060TI(16GB) was just a simple export of the 20 images (VCs) and we have
These are both simple exports of VCs containing identical pairs of AI edits, one a SKY edit and the other a sky selection, with Highlights down at -87 on the Sky preset edit and nothing set on the sky selection!?
Tahoe broke it; PL9.1 worked fine on 26.0 but fails on 26.1
I have a ticket open with support about the masking and the problem with the filters.
1 Like
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
39
I have reinstalled version 9.1 but also updated the Nvidia-driver for my 3060 Ti-card. I got some problems first when downgrading saying that my present database was not going to be read by earlier versions. The installer asked if I would like to install a new database and I answered no so I expected to have to reindex my pictures BUT of some reasons it still can see and work on my pictures after a reboot of the computer.
From what I can see it seems to work perfectly fine again when using the freehand masking without using the premade AI-presets and the exports with a few masks and editing takes around 6 seconds per image again. Images without any freehand AI-masks but with Deep Prime 3 exports in 3 seconds in average (and that is with High Res. Previews and DeepPrime Rendering on.
That’s probably because of DB structure change introduced in PL 9.2/Win to save edits history. BTW, no new problems here with PL9.2/Win compared to PL9.1/Win.
