The Word is out!

DxO PhotoLab
1

image
image1013×404 28.9 KB

https://www.dpreview.com/forums/threads/dxo-photolab-9-system-requirements.4824278/

1 Like
2

Umm, I can’t say it is ‘fun’ and it’s certainly not ‘nimble’ but I can run PL9 and use AI masks (as long as I avoid the preset ones) on my Win 10 box that has a lowly GTX 1050Ti GPU, which has a mere 4 GB VRAM.

2 Likes
3

So I buy a brand mew laptop based on their spec of 6GB VRAM, but Intel Ultra 9 with 8GB VRAM, and that’s not going to work? Now I’m ticked.

4

Requirements have been raised which has promoted my 2019 iMac to junk, at least from a PhotoLab point of view.

Minimum requirements can be hard limits, but my iMac can handle the AI masks, even the "theme” masks as far as I’ve tested them. PhotoLab just takes its time to discover that flower, face, sky etc. and shifting sub-masks between masks sometimes works and sometimes doesn’t.

If there were a switch to hide the AI stuff, I’d flip it. Because there is no such switch, I’ll happily stick to using Lightroom Classic and PhotoLab 8 … or PL9 and the “normal” AI masks, the ones that show up under your pointer.

2 Likes
5

My laptop has a RTX 4050 with 6 GB RAM and it runs PL9 without problems - providing I don’t use the preset AI options, point & click AI selections are fine.

1 Like
6

If I hover over, Ai works fine but fails to execute at times. Sometimes it will fail at export. Sometimes it will go for several images of Ai layered images, sometimes 1 or 2. And I don’t feel safe sending Admin privilaged Zip files full of my laptops data….

7

I wouldn’t send them anything. They should know what the problems are, if they don’t why didn’t they show up in testing. The answer is that the knew and still sent it out.

They are taking the Mickey.

8

So when you bought PL9, you bought the “No AI” version?

If you look closely you‘ll see the masks actually aren’t that great.

9

There is a switch, PL7

10

Yep, its works with 4GB VRAM as you write. Many thanks for you comment, may its helps for forum colleagues. And as its can play nice with 4GB, its can play nice with 6GB, 8GB…

Manual AI mask (selection, area) with 4GB VRAM works fine even starts form PL 9.0 and all afterwards. Just need to know, what is the limit(s) for the ‘play nice’. As ‘Pre-defined AI masks’ is definitely a heavy VRAM user, its a limit for 4GB.

I write a post a while back about that (4GB).

I do this with AMD 4GB, may with nVidia even better.

11

The problem is that for some people it works, for a lot of others it doesn’t. Lots of people with above minimum requirements have crashes and incredibly slow renders and although these have been reported now for weeks……… tumbleweed.

I offered to sell my PL9 update and no-one wants it (although apparently, it’s not transferable :face_with_crossed_out_eyes: ). I tell you what, if anyone wants it, message me and I will give it away. I have seen enough to convince me that this is never going to work properly.

12

My concern about that, what means : ‘8GB was insufficient’. Insufficient for what?

13

At general, i think it’s need to works for everyone (i can speak only for Win, but i think its more-or-less valid also for Mac).

Hard to know ‘for a lot’ why its doesn’t. I don’t blame anyone. But i guess sometimes they simply not ‘follow the steps’ like ‘only Manual Ai mask’ and similar. Sometimes even i forgot the thing. And may some confusing about the whole thing like ‘why its important?’. May some not realize ‘in this times everything use some GPU VRAM’. And some confusion also can happen, like ‘Shared memory’. Also, may some is personal opinion, like 'poor performance". I think it has many-many aspects. But see below what i think about that. May some points not fully related with GPU stuffs, but at general ‘PL9 release quality’ / features.

AI mask related:

  • Seems GPU driver quality vary - at least in the nVidia side (however most of us use NVidia, so its biased a bit) - @BHAYT do this ‘driver version stuffs’ also in his great Export tests (he does incredible work!). This driver quality stuff not good, of course. May some earlier driver more stable, may give a bit better VRAM handling, etc. I already see this ‘driver quality’ even for PL7 vs. AMD drivers - but it was just a few times for me. I guess lot of us not so familiar driver issues at general → and in PL the PL9 that’s the ‘first time’ when its used in amount (other than during Export).

  • May not close all other apps. For me with 4GB, i close everything as possible. Yes, even a web browser in the background can be an issue (its also use some VRAM, may not so much, but an amount)

  • May not follow the ‘steps’ like: ‘Manual AI mask’ only and ‘not Pre-defined AI mask’, ‘DP3 only and not DP XD2s’. May they use Pre-defined masks - as may someone think: 'Pre-defined it just AI what’s can be a difference? What’s can be wrong? - but it has difference. Similar for NR. May some use DeepPrimeRendering also → what also use GPU VRAM - but may someone not realize this (its practically full image NR). Also, ‘Auto-Brush’ now also ‘manual AI mask’ however its may no clear for everyone.

  • Export ‘paralel’ count. If its 1 (one), may less like raise issue like if its 4 (four). For some not see difference on that → “Why not, if its can be setup? What’s can be wrong?” For me with ancient 4GB AMD GPU more than 1 never works even with PL7, PL8, but may even far earlier releases.

  • Willingness to compromise. I can understand this. But even DxO specifications just say like "6GB for AI mask and DP3’. Just say ‘AI mask’, and not say ‘pre-defined mask’, just say DP3 and not DP XD2s. May some not realize this.

  • Very clear the ‘pre-defined’ AI masks (like ‘Sky’) is the more problematic, at general i think most of the issues related with that. → VRAM amount / what you use (like pre-defined AI mask)

  • Pre-defined AI masks detection is complex. May not everyone realize that, as ‘its just AI’ → but for example ‘clothes’ → May get first with AI if somewhere ‘people’ and may ‘clothes’ is need to be ‘around of them’. ‘People’ also need to check for multiple peoples. ‘Animal’ is need to be ‘more animal’ not just like ‘subject’ and so on. So its takes more calculations, may more VRAM etc. And may this complexity / VRAM underestimated for some.

  • Export is a separated process (DopCor)! And for Export photos need to calculate again all the stuffs + calculate NR! And its also use resources! May this ‘calculate’ during Export not fully clear for everyone. Before PL9 AI stuffs ‘only the Export NR use AI/GPU VRAM’.

  • PL8 and all previous use ‘small’ GPU VRAM amount, its was okay for lot of people. After the PL9 AI is a ‘big step’ regarding GPU VRAM usage. May for some it was not clear (the ‘PL8 and below not use too much’)

  • Some say ‘Its not even max out of Memory (boot system RAM and GPU VRAM)’. Yes, off course. Because if it’s cant allocate the needed VRAM amount in one step its cant work. Its not behave like: VRAM Memory usage goes up in gradually ‘slow steps’ like 100MB / 2 second. AI model for masking need to be loaded at once / run only if it’s loaded fully to VRAM (cant be partially loaded). If its can’t allocate ‘instantly’ the required like 1.5-2.5GB, then the ‘Internal error’ raise up. So, that’s the reason if not enough free GPU VRAM → cant ‘max out’ the VRAM.

  • “Shared memory” - No, its cant be used for like an extension of the ‘Dedicated GPU VRAM’ for AI masking. Practically it also means the ‘CPU only’ mode…

  • ‘Stability’ - i don’t think PL9 is ‘unstable’ → if you use what amount of GPU VRAM you has, and GPU driver is good! If you runs out from GPU VRAM → issues raised. But for my point of view its not a really ‘stability issue’. Of course its can be nicer like if its say: ‘Cant allocate enough VRAM’.

  • ‘Sending random photos and related dop is pointless for test’ → No, its not. The .dop is more important, as for check its good to know. Like: step 2: NR only DP3 and not DP XD2s → and if i see the dop its DP3, the this step is okay ->but i see at least one time ‘DP XD2s’… And original photo never a bad idea for test.

  • 6GB. Yes, its may in the ‘edge’. But as its can work in a way with 4GB (what is absolutely in the edge) i think 6GB can be okay. IF you only manual AI mask, DP3, etc. And that’s why ‘optimal’ is 8GB to leave some headroom.

Release ‘quality’ at general

  • Forum has a big bias → as in here usually peoples come with problem :wink:

  • Not easy to compare performance with other Photo apps. Its a ‘pro-and-cons game.’ And not just ‘performance’ compare is hard, but its multi dimension ‘game’, like: image quality, geometry quality, NR quality, user interface, Subscription or not, and so on.

  • Also hard to difference what is ‘important’ regarding performance / stability. For some the export performance is very-very important. For me, i don’t care. I have ancient 4GB GPU, not so fast PC, its just export, takes time as-it-is, I clean up the kitchen in the meantime. Usually i export 10-30 photos, less case may 100 (in one session). Rare crashes with PL is acceptable for me, for others may not - but for me other photo apps rarely also crash. For some has 1000’s of photos in one folder, i guess many of use don’t work this way. And so on.

  • Comparing performance with like Lr. Example: Sky. May in Lr something guessed AI, is not AI (or at least not fully) AI. At least i think ‘Sky’ in Lr not AI (or very minimal). And of course because not AI its ‘fast’. And in PL9 is all AI. Regarding like ‘Sky’ → PL9 quality is quite poor. Yes. And its may raise concern for users ‘performance and quality’. In other hand, i not see too much quality issue on the other pre-defined AI masks types in PL. At least what i check: Peoples, Subject, Background, clothes works okay. May AI ‘hallucinate’ in few cases, but its AI.

  • Also for some this Export ‘calculation’ seems ‘performance issue’, however its just how things works.

  • Export performance* at general ‘as-it-is’. As your PC ‘fast’. As your GPU ‘fast’ → like GPU chip is newer its faster the 8 years old, also GPU clock, GPU memory bandwidth, etc. And of course the photo megapixel count. 40Mpix double the time then 20Mpix. And masking of course, etc. So, like: its takes 10 second or takes 30 sec its near meaningless at least my standpoint of view.

  • Good parts: Manual AI mask (selection, area) is may underestimated. I think its quite good. At least its far better than my expectations. And its not just ‘small selection/Area’ AI, its full AI, like sometimes not just ‘find’ face, but also for hands. Or selection find ‘object’ what was not easy to see with eye.

  • Good parts: Masking improvements like layering, ‘invert shape’ is underestimated - until you not start to use it → Its ‘breath new life’ for all mask types. I use a lot, love it. For some they not interested on that, or not try it. To be fair, its may not easy for everyone, may takes time to get it the potential.

  • Good parts: seems with 12-16GB (or may even less like 10) seems everything runs fine, and fast. Off course, this GB VRAM range the GPU’s quite fasts.

  • Why 'Full preview in progress needs"? Because no real other way to display the thing. And even its need to multiple times like: without correction, with geo correction, with global adjustments, with local adjustments, etc. And also its need to be edit ready, 1:1 quality not just something.

  • As i see, quite a few people think AI mask is saved. Yes, its saved, but all mask (not just AI) saved as ‘position’ and ‘direction’, and ‘exposure value’, ‘AI prompt’ and so on. Just numbers. Not pixel maps!

  • I wonder if someone “I don’t upgrade, because i don’t need this features (and don’t upgrade)”. Okay, i understand that. What can i say. Its also raise a question: what features is needed for wort to ‘upgrade’??? I can understand is like: more better NR, like DP4. Okay, its reasonable. May its happen, may its not. Who knows. May happen in PL10, may its need like more GPU VRAM, etc. And another feature? I wonder what can be?

  • Forum feedbacks, hmm, lot of cases missing. Someone write something issue, and never update what is in the end, like: “Yeah, my AI ‘hallucination’ mask now fine with 9.2.1.” and we don’t have the original photo, so we even cant try it. So, i think lot of ‘i have this and that’ problems may never go too in depth, or not known a follow-up.

That’s all in the nutshell.

Note: i’m aware about Export itself has some issues as others describe very-very deeply. For me not come up, but i accept what others find. However, i don’t think is a deal breaker.

P.s.: as far as i know, Lr also need like 8+GB GPU for all ‘Accelerated’ functions to work.

1 Like
14

Fully agree, no reservations whatsoever!

1 Like
15

Best comment ever! :melting_face:
But yes, I think at general its true the ‘it’s need to be work’. If GPU driver is okay, and if user does/use what can run in the GPU VRAM amount.