If only DxO would join us in THEIR community forum.
I’m sure they could have answered everyone’s concerns in a few simple posts.
Instead they leave us all to make judgements and assumptions based on limited tests where practically everyone’s results are different.
They already have! It is just to read the system requirements!
If you insist on dreaming getting the hell out of even mediocre or sub system requirements it is up to you but don’t blame DXO and Photolab if you are dissatisfied by the outcome.
Once I got a scooter ride in the Judean Hills with a funny American guy. That machine wasn’t all that potent and when I asked him to turn on more gas he screamed to me: “You can’t get the hell out of 8 horsepowers”.
After that at least I know these facts! … and that is a general rule that can be used in most similar cases of everyday life. If that rule was more widely embraced discussions like this would never occur!
Except other software that does the same sort of thing runs better on such systems during the editing phase (not counting export times).
If we found that e.g. Capture One, Lightroom, Affinity, Darktable etc. all chugged then - yes - we’re clearly being ambitious with what we hope older systems can achieve.
Although speaking of export times…
I just had a fun experience with PL10 where - for no reason at all - export times with DeepPrime 3 rocketed to over 30 seconds per image. Fortunately restarting fixed it and I was back down to 10 seconds per image in some cases (which is great for my system, no complaints there).
It just goes to show how PhotoLab still has quirks and performance issues.
We can compare Capture One with Photolab and stress that Capture One makes it with 8GB VRAM but we can also see very clearly in some tests that the masking accuracy is far inferior to PL 10. Neither C1 or LR to my knowledge can denoise even my 4K screen with high res previews like Photolab 10 can today. To use that takes a good graphics card with 16 GB VRAM. If you haven’t one of those get one. If you don’t feel for spending the money it takes don’t use Photolab. Stick with C1 or LR instead. You can’t get it all for 50 Crowns as some people say in my language or “Man kan inte få allt för 50 spänn”.
Here is the lists!
The question IS NOT if Photolab 10 works as good as Lightroom or C1 on “your” or the tread owners computer.
The question is what is actually recommended for a smoth operation.
If you can´t meet these specifications you don´t ought to choose Photolab 10.
It doesn´t help to have a good cry here I´m afraid. That is just a waste of time for all of us.
| Minimum | Recommended | |
|---|---|---|
| CPU | Intel Core 10000-series / Core Ultra 7 165H / AMD Ryzen, 4 cores | Same families, 8 cores |
| RAM | 16 GB | 32 GB |
| Disk | 30 GB free | 50 GB free |
| Display | 1280 × 768 | 1920 × 1080 |
| OS | Windows 10 22H2 | Windows 11 24H2 |
| AI Masks | Windows 11 24H2 required | Windows 11 24H2 |
| NVIDIA for AI | RTX 3000 series, 8 GB VRAM | RTX 4060 Ti, 16 GB VRAM |
| AMD for AI | RX 6000 series, 8 GB VRAM | RX 7700 XT, DxO lists 12/16 GB depending on current page revision |
| Intel GPU | Arc, 8 GB VRAM | — |
| NPU | Intel AI Boost supported | — |
DxO specifically says latest GPU drivers. DxO User Guides
One rather striking change is that 16 GB VRAM is now DxO’s recommended NVIDIA configuration, not merely something useful for enthusiasts. Their reference GPU is an RTX 4060 Ti 16 GB.
Minimum is 16 GB RAM, 30 GB disk space and macOS 15.7 Sequoia. DeepPRIME 3/XD3 requires at least Apple M1, or an Intel Mac with Radeon Pro RX5700. AI Mask and dust detection require Apple M1.
Recommended is M2 Pro, 32 GB RAM, 50 GB free disk space and macOS 26 Tahoe. DxO User Guides
And PhotoLab 10 contains a new AI memory management option that can unload AI models from GPU memory when necessary; DxO explicitly says this is intended mainly for configurations close to minimum requirements. With 16 GB VRAM, you normally shouldn’t need it.
It was really smooth with 3090ti (alone no dedicated graphic card) so 24 GB Vram. I suppose vram capacity is more important than gen 3, 4 or 5 since this was (trial has stopped) smooth when working with ALL preview option enabled (so Final image in preview with DeepprimeXD3 on and all lens corrections). Maybe my cpu plays a role here since it significantly helps my graphic card during intensive workloads, specifically in multi-GPU coordination - not relevant here - and data preprocessing.
I am the bad duck here. I never do that but extensively test what I’ve got and do not change if it works fine. I do not want to disrupt other graphic softwares which work like a charm and discover this in production without understanding what happens. And I think that’s a good advice for professionals.
I upgrade only when needed. And I know drivers I have don’t work well on some stations. Not mine.
@JoPoV Well, concerning our experiences from version 9 I can understand your way of not fixing anything that isn´t broken or works fine already. When I started to use version 9 I had to go back 20 versions!!! (yes Nvidia do a lot of versions:- ) ) in order to get it working at least with the exception of not using the premade AI-models in the menu.
I read Photolab consume more systrem resources than when using the hoovering over medthod. The first method has to analyze the whole picture and the later just the are a you are pointing at.
The Ti cards of the 30-series are faster and better on parallel-processing than the non-Ti cards. So a 3060 Ti 8 GB card works better than a 3060 with 12 GB with Photolab 9 and presumably with 10 too. You have more of both of these critical resources, so congratulations 
I understand your frustration, but you maybe could moderate the title of this topic, since It doesn’t fully reflect reality and remains at the top of the list throughout the launch of Photolab 10. That’s not exactly brilliant for a launch. (Unless this is your intend …).
For me, the main issues (and I could really go into this – but I won’t here) mainly concern large volumes that need to be processed quickly, where there are numerous shortcomings and problems, but it seems that this doesn’t bother you even though you’re in this field.
My trial of PL10 went very smoothly without the problems you mention, so I find this unfair. And I have a professional’s perspective on the tools I use. Photolab is not for all professionnals, but suits the need of those who want quality and aren’t bothered by certain issues.
I have absolutely no connection or contact with the DxO team (and if you read my posts on the forum, you’ll see that I don’t mince my words when it comes to things that aren’t right).
It’s up to you to decide.
With my 5070 Ti 16GB card I really think PL 10 is really nice and fast to work with. As an exaple it also fixes to handle smart masking (using the menus with the premade AI-models) on a lot of pictures. Then the AI-model decides where to put the masks on for example a face or a person in all that sedt of pictures dynamically insted of using that mask statically instead which is not a good idea.
This method did not work for me earlier when I had myu older 3060 Ti card with just 8GB.
So I don´t think complain over problems I had earlier with my 3060 Ti card is really fair NOW when DXO has increased the recommendations.
I think people shall read what they are telling us in the specs before starting to complaing here and creating all sorts of confusion. Complains like those are right out irrelevant and ignoring that info is just to ask for problems deserved.
I have no doubt about this. You are in the powerful area.
I think 8GB is the problem. not 3xxx serie.
My son has the card type you have too and I don´t think he has any problems either, so it might be a slightly cheaper option I guess. The 5070 Ti cards I think has increased in price around 60% here since Oktober, so I understand building a computer powerful enough for this can be a challenge for many.
3090ti is still a good graphic card for photolab :
| Feature | RTX 5070 Ti | RTX 3090 Ti |
|---|---|---|
| Architecture | Blackwell (GB203) | Ampere (GA102) |
| Process | 5 nm (TSMC) | 8 nm (Samsung) |
| CUDA Cores | 8,960 | 10,752 |
| RT Cores | 70 (4th gen) | 84 (2nd gen) |
| Tensor Cores | 280 (5th gen) | 336 (3rd gen) |
| VRAM | 16 GB GDDR7 | 24 GB GDDR6X |
| Memory Speed | 28 Gbps | 21 Gbps |
| Memory Bus | 256-bit | 384-bit |
| Bandwidth | 896 GB/s | 1,008 GB/s |
| L2 Cache | 64 MB | 6 MB |
| Boost Clock | ~2,452 MHz | ~1,860 MHz |
| TDP | 300 W | 450 W |
| Interface | PCIe 5.0 | PCIe 4.0 |
| DLSS | DLSS 4 (Multi Frame Gen) | DLSS 2 |
| FP8 / FP4 |  |
 |
| NVLink | |
|
| Outputs | 3× DP 2.1 + 1× HDMI 2.1a | 3× DP 1.4a + 1× HDMI 2.1 |
| Release | January 2025 | March 2022 |
Waiting for gen6 to upgrade my gen3 stations. If everything doesn’t go pear-shaped before then.
I wholeheartedly disagree.
A similar pattern has been observed in poor games developers: not bothering to effectively optimise their work (instead, often relying on DLSS and frame-generation to do the work for them).
I don’t need, or want, PhotoLab to render de-noising and sharpening in real time as I edit, it’s extra work that takes place with every single edit and that puts extra, unnecessary strain on our systems.
The only way to turn it off is to not use those tools until the very end of your workflow or implement partial presets but, as we’re seeing in another thread, PhotoLab doesn’t have partial presets in a particularly logical and accessible fashion. The alternative is to enable these tools at the end of a workflow but that can only be done per-picture or across a set (and hope you don’t accidentally change something else on every one of your shots while they’re all selected).
This is not good user design, but I digress.
Writing ever-increasing minimum requirements is not an excuse for less effort in the optimisation dept. and evidently the competition is ahead there. Glibly saying “just buy better hardware” (another statement parroted by unhelpful teenagers in the gaming community) is no real solution; right now we’re seeing hardware prices at an incredible high, and don’t want to remortgage for a top spec machine.
Unpleasant, boot-licking attitude aside, you’re wrong there too.
If there’s no feedback that the software isn’t well optimised vs. competition then the status quo becomes “all is fine”, when it clearly isn’t, and there’s work to be done on the optimisation side of things.
You’re not obliged to chime in, so feel free not to “waste your time” further. Repeating yourself would be a waste of our time.
As I mentioned earlier, you don’t necessarily need a high-end computer to work effectively with PhotoLab 10.
On my underpowered laptop (AMD Ryzen 7 5700U without a graphics card) crop takes less than 4 seconds to be ready, the AI-masks does too (45Mb Z8 files). Only when an image contains many AI masks does the initial processing time (calculating Preview) increase noticeably.
While editing, I spend less time waiting for Photolab than I do evaluating the image and deciding on the necessary steps myself.
No not at all. Are you prepared to pour gasoline into a car with a diesel engine because it happens to be a lot cheaper. Here diesel can cost say 24 SEK and gasoline around 17 last time i checked. A difference between 5-7 SEK has been the case a long time. There is roughly 10 SEK to a U$.
Yes a hard guy might say a mans got to do what a mans got to do - guys guessing - instead of boot licking and it might be what explains why some write upset posts here when they have reached the end of their long and winding guessing trial and errors in the middle of nowhere.
Why not learn a few things from your girlfrind or the closest female there is. Women rarely makes as stupid mistakes as guys guessing men because they read the instructions and recommendations and are not stupid enough to neglect good advises before they start paragliding or kite surfing or using new versions or types of softwares they haven’t used before.
Here a picture that illustrates this very clearly,
This looks like an gearbox from a Volkswagen
I remember another story, also with an old Volkswagen, when I was working at a General Motors import plant in Stockholm during my summer break from the school, where they also had a “Do it yourself” -workshop. There a guy had put a car on a lift to do a few things - one was to grease up the breakwires. So he took a grease gun and started - and he did not stop before he had filled the whole rear break “drum” with grease.
After that he had to leave the car to the professionals at that station to fix the problem. This little incident would probably not have happened if his girlfriend had done the job instead.
If not, prepare yourself. Wont be good gazoline if any in not too long (I’m talking about cars !).
My apology! Sorry for my English. It is not my primary language - is it OK if we switch to Swedish??
Det är helt OK för mig. Då kan du bara använda Google Translate och översätta det till Engelska om din svenska är lite rostig!
Possibly a poor analogy, as I changed the leaking radiator coolant pipe on my car not so long ago, while my partner wouldn’t know where to start with such things 
(She’d be the first to admit it, cars are not her thing - they are mine)
FWIW I don’t take joy in saying all these things. But over the years I (and others) have identified some strange, silly, or utterly stupid design choices by DxO that - if fixed - could help their software run better on a wider range of machines (more happy customers for them - yay!).
The stuff where their software excels - DeNoising and sharpening algorithms - are not where the silly choices happen. They work as expected and they work on my machine, right now.
That’s why it’s frustrating to see them not address the optimisation elsewhere, as other companies are ahead of them and they don’t seem to care. They say “Just increase the minimum specs” and that’s lazy programming.
