Of course I export the images I work on. How do you think I notice the rendering has changed? 
For me the latest version of PL9 performs worse than any previous version, with one photo export taking over 2 and a half minutes.
My hardware is not faulty and this problem (and other severe performance issues where the whole program locks up for 10/20/30 seconds at a time while it “thinks”) exists with the latest version of DxO’s software, and no other.
This is a software/optimisation/performance problem, not a system problem.
1 Like
I applied the AI people mask on a 8Mb jpg file adjusted the exposure and exported the file as a jpg. The Export took 58sec.
I remember the DXO product managers at the launch of PL7 or 8 saying how they don’t follow the competition and only introduce features that out perform the competition (paraphrasing). I hope they have learnt some humility.
1 Like
It that sense yes but I don’t think that’s the intention. More add denoise but then mask a subject and go a bit further with the subject.
This short video shows this…
There may be several reasons for this, like:
- Most probably your GPU is not used for exports. Check your ‘Performance → AI Acceleration’ and ‘Enable OpenCL’ settings. If both are OK, check DxO.PhotoLab.txt log. If GPU was used, something similar to the following should be there (applies to PL9.2.1/Win):
DxOCorrectionEngine - Info | InferenceDevice: selected inference device is NVIDIA GeForce RTX 4070 (11.7 GiB, type=Discrete GPU, id=0x118a7)
DxOCorrectionEngine - Info | Using adapter NVIDIA GeForce RTX 4070
If it’s not used, you’ll probably see something like
DxOCorrectionEngine - Info | Enter GetWARPAdapter
DxOCorrectionEngine - Info | Creating DXGIFactory1
DxOCorrectionEngine - Info | DXGIFactory1 created
- You are short of some resources like RAM, virtual memory (check Commited memory in TaskManager and swap config – should be Auto), GPU VRAM, or Shared GPU memory. Check your CPU usage when idle.
- GPU and/or CPU are not working at full speed. For GPU you may use the following command (nvidia-smi comes with the driver):
nvidia-smi -l 1 --query-gpu timestamp,pstate,memory.used,memory.free,utilization.gpu,temperature.gpu,power.draw,clocks.sm,clocks.mem,clocks.video,clocks_event_reasons.active
Normally PState should be 0 during the export, perhaps for DP3 going down to P3 rarely. Check clocks if they are near max. The clocks_event_reason.active flag should be 0 most of the time, with few 1 or 4 possible. If you see anything else, you probably have a hardware problem. Verify cooling and power supply. Check Windows Power Management settings. BIOS/UEFI is a longer story…
4. Some other reason, which may include bugs in anti-virus, corrupted installation, faulty drivers, etc.
5. Your hardware simply lacks power.
In my case it takes less than 3 minutes for PL9.2.1/Win to export 100 x 45mpx uncropped Z8 raws with DeepPRIME3 denoising and a large set of edits used (all optical corrections, SmartLighting, Clearview,Contrast, SelectiveTones, LSO, …). With ‘Use CPU only’ it takes about 90 sec to process a single photo.
Btw, running exports with PL window minimized makes it about twice slower in my case (i7-14700KF + RTX4070 + Win11 24H2, 26100.7171). That’s because Windows scheduler treats “minimized” processes differently.
I have seen this behaviour and my system is using my Nvidia 2070, an export with no subject based AI mask takes 20sec with subject AI mask 2min 39sec.
I posted above where a jpg that I applied a AI Person mask to, export took 58sec.
It’s laughable performance.
1 Like
That’s probably due to VRAM allocation issue, well known and described in this forum by several people. We’ll have to wait for fixes from Microsoft/NVIDIA or a workaround from DxO, if that’s possible.
Watched the video. I love the way he says “and there you go, you can clearly see the difference”. This is a technique used in sales called “the presumptive close”. It is based on the belief that no-one will turn around and say “Bollocks”. Take the bird. 25000 ISO - an extreme beyond extremes. And yet there was virtually no difference between denoising the whole photo and then after, denoising the bird as well. Literally no difference and indeed, the bird looked better the first time round.
I get that that is the way it is meant to work, it just doesn’t really, does it? See what I did there
The AI (let’s call it AI, but it’s not) isn’t ready for public consumption. We are the Beta testers.
@Gareth More like Alpha testers if we are being honest, notwithstanding the issues related to the Nvidia drivers, which still should have been tested and resolved before the product was released, instead we got a premature release and (some of) the problems still continue 3 months later.
But apart from the damage caused to DxO’s reputation and the frustration generated in the userbase was there any other/better way to actually to expose the new AI features, in particular, to a wider number of users to discover and, hopefully, iron out the wrinkles!?
I am not sure quite what you fail to understand.
A new updated software comes out and users are experiencing problems, to differing degrees but problems nevertheless.
You premise is that they need to adjust their hardware and if that doesn’t work then the hardware must be faulty. There is a major flaw with your thought process. There’s not much thought and little process.
When you have two constants, let’s take PL7 (or 8 perhaps) and the computer. And this works, it works well. If you changed your computer and the software didn’t work, then you would look to see what the problem was with the computer.
In this case, the computer is the same but the software is different and yet you believe that the problem lies with the computer.
The fact is that PL9 does not work very well without users making significant upgrades and way beyond DxO’s suggestions. Now that investment maybe worthwhile to many people if PL9 worked well but the AI mask is not very good, it’s not very good at all.
Unfortunately, DxO are dropping the ball. But apparently it is the ball’s fault for being round and if we had a square ball things would be fine.
1 Like
If you want to use “AI” Masking as the USP, you better make sure you get it right. It’s not. Think about it like this. The Airbus A380 carries 850 people. It is basically a plane that is run by software. Imagine Airbus built a plane, got a pilot, piled on 850 people and took off. A few hours later, the plane falls out of the sky. Airbus turn round and say “well there’s bound to be a few teething problems”.
Clearly for you there may be very little added benefit. However, I have owned every version of PhotoLab since PL 1 and for me PL 9 has very significantly improved functionality over PL 8 even without using the AI masks.. PL 9 also runs well on my 2016 Windows 10 machine which had been updated last year with an Nvidia RTX 4060 graphics card and a WD Black 2 TB SSD drive.
Exporting with the various DeepPrime versions is actually faster on PL 9 than it was on PL 8 and PL 7. . Exporting files with DeepPrime 3 averages 6 to 7 seconds. Exporting files with DeepPrime XD2s averages 9 to 10 seconds. That is without an AI mask selection. With the inclusion of a moderate amount of AI masking the export time increases significantly. With DP3 the average export is then around 13 to 15 seconds, and with XD2s It’s around 18 to 20 seconds. In other words, exporting with DeepPrime and AI masking approximately doubles the export time. In any case, I can comfortably live with these export times.
Finally, using the latest Nvidia driver with PhotoLab 9.2.1, I have not had an internal errors or any other significant issues for the last several weeks, and I use PhotoLab every single day.. Snce it runs so well with my very old processor, an i7-6700 @3.40, with a relatively new but now somewhat middle of the road graphics card, and only 24 GB of RAM of 10-year-old ram, , I wonder why I have no significant performance issues while others with much higher spec’d PCs are having difficulties running PL.9.
Mark
Well I see that apparently all you what to do is criticise and whinge.
I was only highlighting the fact that maybe if that was the only way you were utilising the local masking of denoise then maybe you weren’t using it as DxO had intended/envisioned it would be used and offered a video demonstrating its use.
Was the bird night and day different? No. I wouldn’t expect it to be tbh. My understanding is it’s to add a little bit more over and above the denoising done to the whole image.
However it’s somewhat disingenuous to say the example with the bottle showed no change. Again. Small and subtle but to me noticeable even on a small phone screen.
Maybe that’s enough for the intended use case and it’s a use case you see no benefit to. Fine move on.
However. Instead of going maybe I’ll fire up PhotoLab and give that a try and see what I think you jumped straight back on here to scream into the void.
I get it. You are not happy but at which point in the trial did you realise it didn’t meet you expectations and decide to skip version 9 or are you having buyers remorse?
I’m no fan of DxO/PL. it’s a tool that works for me.
I was happily using a combination of PL5 and LR but a change in computer means I can no longer install PL5 and a change in attitude and a lack of desire for a subscription mean I no longer want LR.
I trialled PL9. It does what I need still and does it well.
Could the so called AI mask be better? Sure. LR/Adobe clearly wins here but I personally can get the results I want or close enough to be happy still based off my trialling of PL9.
Will they get better though? Hopefully but again personally I didn’t buy it for those so called AI masks.
Anyway. Yes. It should all work as ‘promised’ and unfortunately it seems it’s not for some and yes I agree DxO need to sort it out asap for those that are having issues. Communication could be better here from DxO but I seem to recall it could have been better back when I bought PL5 and later trialled 6 or maybe 7 too.
2 Likes
Clearly no ! When you use IA , the problems arrive. Software closing alone . Suddenly exportation takink a long time…
My config is OK : I7 ultra 32 Go , RTX 5060 8 GO
2 Likes
It’s really not my fault that you think it is a hardware fault when the only thing that has changed is the software. You could have just said “Yep, that’s a good point. Maybe we should be looking a bit harder at software requirements and the limitations of the new software. Shooting the messenger is never a good idea.
As for buyers remorse. I updated PL7. PL7 worked, everything in PL7 worked from the get go. PL9…….. not so much. If we ignore the AI mask (recommended) then is PL9 much better than PL7………………… not much. I like the DP for X-Trans, everything else is meh.
The point is………….. A long time ago there was a word processing software package called Word Perfect by Corel (if my memory serves me correctly). They used to ask for feedback. Well (I forget the edition number) but a new version came out and was half baked. The owners could see MS Word in their wing mirrors and they rushed it. Then they decided that they were getting so much “feedback” that they wouldn’t read any more emails on their “feedback” email address. So they left it. Most people don’t even remember the top word processing package in 1991.
When your customers start to tell you there is a problem. Either your customers are wrong, or you are. Don’t blame the messenger.
I had a similar impression when I first used this version.
FInally I discovered a trick 
In the settings—>performance page, instead of using “auto” option, I used “nvidia”
In other worlds I force PL9 into using the GPU, instead of letting him decide.
It works. Performances are back to normal, or even better then it used to.
1 Like
And where exactly did I say that then?
So, more than 8Gb of VRAM is needed to export a jpg faster than 58sec? As I said, laughable.
Not sure why you’re getting 58 second exports. As I indicated in an earlier post I have an Nvidia RTX 4060 with 8 GB in an otherwise very old Windows 10 machine from 2016.
Exports are very fast for me. Perhaps you’re processing larger raw files than I am which would definitely have a negative impact on export time, or perhaps an older graphics card. My raw files are from my 24.5mp Nikon Z F. If you have, for instance, a 45 mp full frame sensor in your camera, I could see DeepPRIME and a modest amount of AI masking taking that long, especially If you have a medium level card like my RTX 4060.
Mark