Have you raised a case with DXO?
Seriously @Johnaalex , no more dribbling around now please, this isn’t a game of ice hockey
You are speaking for DXO as I have understood it, so now it is really high time finally for you or somebody else at DXO that still has some ears TO LISTEN too. You have to go back now to DXO’s R&D and just ask them to walk through the relevant treads here at their own companys user forum in order to get in synch with their customers/users reality. … and if you ain’t the right man for the task please bring some others that are and are able to take this more seriously.
DXO has to focus on the malfunctioning AI-models in Photolab and nothing else until this is sorted. I hope you realize now that this is nothing you can handle with just your attempts of verbal damage control. This will take same serious analyzes and actions from your R&D and not just even thicker smoke screens.
1 Like
Sad to say that even with the new NVIDIA driver installed I am still getting an internal error when doing an export! 
1 Like
I wonder why? Even with my old RTX 2060 6GB GPU everything works fine, but slow.
2 Likes
I write a few times (may even for you) a few things about VRAM usage. I’m on move cant copy that. Anyhow, please try the followings:
Preferences - Deepprime rendering - turn off
Preferences - Paralel exporting or what change to 1
Used DP3 NR
And you may in the safe side.
Please let us to know.
No I´m not. The first workes with three masks made with Sky, Person and Vehicle. The second crashed. I have just one parallell process set. Have no DeepPrime rendering of previews. It is really hit and miss exporting them. On top of that much slower.
When I masked on freehand the exact same picture doing the same but with freehand masks it works to export.
This picture and three simple masks: 47 seconds with your premade AI-presetrs and 8 seconds with the other much more efficient method. Almost 6 times faster. Do you understand me now when I say that you have some optimizing to do? You can´t check out from this case yet.
Please download Capture One trial and see for yourselves. In C1 we have never had these kind of problems despite GPU-cards with just 6 or 8 GB. There lies your competition barr.
2 Likes
What is means exactly (as ask to be sure)? Its crash only in ‘Export’? Or its crash when its displayed (‘rendering preview’)? Or sometimes crash in each?
DP3 or DP XD2s for export?
What happen with only one (1) ‘Keyword AI mask’ (pre-made preset)? Like only the ‘sky’ mask enabled and the ‘person’, ‘vehicle’ mask disabled? What happen i sky + person enabled?
Add-on: You export 1 and 1 (separately) or in batch (2 photo selected and export in one batch)? 1 by 1 change anything?
Do you understand me now when I say that you have some optimizing to do? You can´t check out from this case yet.
I understand all. I do a calculations, measurements, ‘reverse-engineering’ in the last weeks, may some of it helps out for a some of us. For example yesterday i comment about how ‘simple manual AI mask’ calculation times, and its seems more-or-less linear time (see below)
Yes, it’s DxO job to make things better and faster.
However, i try to concentrate how now (as today) we can get something “limit” where editing, exporting works without crash.
May some ideas works out, may not. May we know something more about PL behaviour.
I have 4GB GPU, and not a too much chance (budget) to get better GPU (more VRAM) for the next half a year. And i try to seek what is the max i can, and may some of my measurement/experiments/ideas may helps others.
Note: I not concentrate on export times - as it’s practically based only in GPU performance, and AI Engine performance (AI mask count / AI mask types (like ‘sky’) - and of course GPU VRAM amount (as this one the cause the crash - at least i’m pretty sure on that)
Anyhow, you give a few ideas to me about testing a bit deeper the ‘AI keyword masks’. Thanks’
Before we got Photolab 9.1 and the last Nvidia-driver (581.57) I could not really even try to apply the premade AI-presets. It crashed instantly when Photolab opened and began to render the previews BUT I had no problems with using the frehand method of applying the masks, even with the earlier versions of Photolab (9.0 9.01) when I downgraded the driver to version 572.83.
So, your recommendation is NOT for me to set parallell processing to 1 or not using either High Res-previews or Deep Prime-rendering of previews. All of that works fine if I just do not use the premade AI-models in the menu and instead use the freehand methods when applying masks. I have no problem what so ever with either scrolling, exporting ot printing. Those problems ONLY occur when I use the the premade AI-presets. It is really that simple and that is why I have asked representativs fron DXO to inform their R&D about that - probably they ought to know that by themselves by now too.
Using the premade AI-models in the menues destabilizes the system and makes it 5-6 times slower than with the other method. From what I have read it looks like people are still banging their heads against “the wall” insisting to use the premade AI-models. I have stopped doing that since long and I can honestly say that I dont have any problems at all. If I stick to that workflow Photolab now works fantastically well and is very very efficient - it is even faster to export my pictures with AI-masks and Deep Prime 3 than Photolab 8 was without any masks!
My average exporttime when exporting 200 pictures with a mixture of both traditionally editings and using the frehand method were just 6 seconds the other day and I´m totally happy with that. With version 8 and Deep Prime XD2s the average used to clock 7 seconds.
Despite that even I like to be able to use the premade AI-models because I can´t use the other method to propagate the masking settings from one picture to a set of other similar pictures. Even I would have liked that since that can save a lot of time for us all.
1 Like
May also for you. As you write ‘The second crashed.’. The ‘crash’ why i write.
Actually I’m pretty sure most of the ‘crash’ (if not near all) related with GPU VRAM usage or some GPU driver (but i think the later in the case of NVidia now seems solved)
As i see, both paralel processing (in export with DP3 or DP XD2s) and deepprime preview consume some GPU VRAM (as its also use DP3 or DP XD2s), and also DP XD2s use a bit more GPU VRAM than DP3. That’s why i write this → may saving some VRAM helps you out on ‘crash’
I wondering about ‘High Res-previews’, may i not mention in here (at least in previous comment). Anyhow, i don’t think ‘high res preview’ does nothing special about GPU VRAM (and even performance is quite good).
I not write about export times or whatever - actually i don’t care to much about that, as its as-it-is at the moment, if not crash.
I have tried all these things and changed 2 to one and even turned off both High Res and DeepPrime rendering of previews and that did not protect me from crashes when exporting. The only thing that gives me stability is really to completely avoid the premade AI-masks and use the freehand method of point and click instead and for me it is a no brainer to do so if I shall get something what so ever done.
To use the premade AI-models is really bordering severe self-harming for many users. Why choose to wait five times plus time extra for an export to finish if you don’t really need to.
I will probably retire my present computer and buy one with an Nvidia 5070 Ti 16 GB GPU and a bigger SSD instead (have just 1 TB now and that is too small really) in the end of November during Black Week. There seems to be a shift now from the 40XX GPU-cards to 50XX in gaming stationary computers and maybe I will assemble one by myself. It will depend on what is available at the time. I think the gaming PC: s often has been price worthy options even for non-gamers. Now both Photolab and local AI in general for example running Gemma 3 on LM Studio will really fully use even 16 GB. Now i only can run Gemma 3 4b when having an 8 GB 3060 Ti.
With Gemma 3 12b we need 16 GB and I need the 12b in order to solve the need for the info resolution I need with iMatch Autotagger too is I want to run AI locally and not being dependant on American OpenAI. Google Gemma 3 is free on LM Studio together with hundreds of other interesting AI-models.
2 Likes
I think at this point (with the likes of PhotoLab) we really need a hybrid gaming PC (that is: a gaming PC with robust CPU, GPU, and RAM… …that also has substantial storage etc. such as a photographer might need. A gamer could get by with far less).
Fortunately I still enjoy gaming from time to time so both suits me well, but it’s funny that at this point it’s not the latest games that are making me think I finally need to upgrade, it’s PhotoLab 9.
1 Like
That’s exactly how I had my PC built 10 years ago.
In the store: “Build me the fastest gaming PC with the current hardware that money can buy, but scrap all the blinking coloured RGB LED bullsh*it that gamers love”.
Apart from a graphics card upgrade in the meanwhile, it still runs strong.
2 Likes
That’s exactly the identicle message I got from DXO support. The first time I tried after installing the new driver it failed on the images that failed before. However, having deleted those images and reinstalled the images and renaming, all seemed to go well. So far since all seems to be working well with the AI masks.
1 Like
Based on DxO’s obsession with pushing us all onto the latest macOS, my guess is that they are constantly using the latest M1 Ultra, M2 Ultra or M3 Ultra. Perhaps some of them get the M4 Max (almost as fast as the M2 Ultra for most tasks). Since I like stability across a wide range of applications, I’m as far behind as I can be in terms of OS. Right now that’s macOS 14 Sonoma.
On M1/M2 Ultra, there are no performance issues in PhotoLab 8. Performance is just fine on an M3 MacBook as well or an M2 Mini Pro or an M4 Mini. Unless you’re married to Windows, if you’re serious about PhotoLab it’s worth getting an Apple Silicon computer just to run PhotoLab. The one issue is that Apple charges €400 for the memory upgrade from 16 to 32GB of RAM. The least amount of memory I’d like for an Apple computer running PhotoLab is 32GB RAM.
It’s not that PhotoLab alone requires the 32GB RAM but by the time one is running PhotoLab, FastRawViewer, ImageOptim, a web browser and a music player, maybe a pixel editor Affinity Photo, 16GB is just not enough. If you’re willing to open and close applications all the time, 16GB would be just enough.
So effectively PhotoLab costs €1359 not €259. PhotoLab at €259 + €1199 for a Mac Mini M4 with 32GB RAM and 256GB of storage (buy an external NVME hard drive and case and boot off of that for 4TB startup drive for €300, plenty for PhotoLab and even video editing, buying that disk space from Apple would cost €2000).
2 Likes
THAT is the crucial point:
How much money am I willing / able to spend just to run a certain application (regardless of whether it’s on Mac or Windows) ?
4 Likes
Seconded/thirded.
It’s a curious thing when I can run various games and other intensive applications well enough on my PC, but PhotoLab is the thing that makes it slow to a crawl, even when other similar applications do not.
Editing photos is not like rendering a video file - it’s far more interactive and the application needs to be “snappy”, not “go away and make a cup of tea while it has a think”.
never ever get a computer with minimum RAM, because that’s what the computer itself need to run (at least), 16 RAM is for people doing minimal work on computer, 32+ RAM is what people need if they run photo/video software.
people can argue with me on this but at the end of the day, i don’t have time to go make another coffee while the image or video is being processed.
1 Like
Does not seem to be the case with Lr
Back when I still had a subscription to Adobe Lightroom and Photoshop, my M1 MacBook Air with 16GB memory often spilled over into swap space (where the internal SSD is used to supply additional virtual memory). Same situation when I began using DxO PhotoLab 8. My MacBook ran fine. Apple does a good job of managing a heavy software load, but SSDs are funny devices and get used up in a different way than older spinning-platter hard drives do. I did not want to continue using the internal SSD for extra memory.
So when I moved up to a new M4 Mac mini desktop, I paid for 32GB of memory. Now I can throw anything at my Mac and still have memory to spare.
1 Like
My Lightroom, Bridge, Fileloupe, FastRawViewer, ApolloOne, Adobe Camera Raw, and Photoshop memory usage routinely spilled over into swap space with only 16GB.
1 Like