As you can see, neither meet the min requirements for GPU and have old drivers, however…
I processed 28 photos each with several AI masks and sub-masks with Deep Prime 3.
I find AI “add an area” to be almost instantaneous and very accurate - far faster than the AI presets. Usually, the subject is masked properly without needing a touch-up. In photos with multiple subjects, I drew an area around each - very fast and accurate. Far faster than Topaz Sharpen AI - which I did not use on these photos.
Creating multiple masks each with multiple layers is fast and trouble free.
Really, this new version operates almost as fast as PL7 which is on the same computer.
I could have fine tuned the hair a bit but the photos are good enough.
What I did with most of the photos was select the subject and invert the mask, then darken the background a bit. This is a lot faster than using the background preset.
The export time was 5 min 50 sec for 28 photos for an average time of 12.5 seconds - slightly slower than PL7 without AI masks.
No crashes even with Photoshop CS6 running at the same time.
The Office computer is slightly slower with an exporting time of about 20 seconds.
@Allan Ignoring DxOs recommendations for a moment I have been doing tests with some Nikon images and found that AI takes around 4GB (slightly less) and exporting then pushes my 2060(6GB) card over the limit so I am surprised that you have been so successful!?
DP3 seems to take even more GPU memory than XD2s so how you are getting through an exporting run with AI is difficult to understand!
I have been successful but only with my 3060(12GB) card.
The drivers you are using are very old which may be a clue to why you have squeezed them into a 6GB GPU , the GTX1060. What is the memory for the GTX1660 card?
Would you be able to run GPU-Z while doing an export run and take a snapshot of the memory usage?
Plus do you have an image with your edits that you could provide so that I can test, although my 6GB 2060 will soon be out of my 5900X when a new GPU arrives later today!?
I am managing to run PL9 on my PC - Windows 11 Pro, i7-10700F CPU @ 2.90GHz, 32GB RAM and a NVIDIA GeForce GTX 1660 Super 6GB GPU with Studio Driver 577.00
I am currently only processing single Nikon 45MP raw files, via the Lightroom Classic plugin. I am not experiencing any unexpected issues, including applying AI masks or perspective correction.
My PC does not meet the minimum GPU requirements, and whilst performance cannot be described as quick, I am not getting the ‘Internal error (Correction failed on the Execute stage)’ at this time. However if I update my GPU, with the latest studio drivers 581.29, I experience this issue.
One thing I have noticed is my export times (100% jpeg) back to Lightroom Classic, via the plugin have become extremely slow.
@Allan Thank you for your prompt response and I apologise for not making a prompt reply. I did attempt to test late last night and discovered you are using a DCP Curve, Adobe DCP Curve to be precise, so PL9 will do nothing with the image on my system and I haven’t used DCP Curve so a little assistance would be good.
I guessed that the other GPU would be 6GB and your GPU-Z log showed it rising to maximum but you are “lucky” to get away without problems!
The Nikon images I was using are from the images on the Google spreadsheet for GPU performance logging. My own camera of choice is a Panasonic (Lumix) G9 with an Olympus 12-200 (24-400) attached, I also have a GX9 and an EM1 Mkii (which is where the 12-200 lens came from) so I can try images from my own cameras as well.
Hi Bryan
Have you played around with Allan’s image and your GTX 2060 6GB GPU?
I also imagine that the relatively small size won’t cause any problems.
Wolfgang
@Wolfgang No and yes and no. No not on my 2060(6GB). yes on my 50506Ti(16GB) and no because it has a DCP Curve and I don’t/haven’t used them. So I got various problems with the image on the 5060Ti(16GB) i.e .with a batch of 40 copies _1image + 39VCs)
I fitted the 5060Ti(16) immediately it arrived yesterday afternoon, before I got the reply from Allan. I am currently running on the latest drivers which seemed to work but things happened I didn’t expect, this is the run with 2 copies, as before, but memory usage shot up 10821
With 3 workers it went even higher and with 4 workers it failed, i.e. it had exceeded the memory capacity of the 16GB card.
It looks as if with the latest drivers PL9 is adding an AI component in VRAM for each worker if there is space(!?) but then tries to process when the additional one wouldn’t fit but with the 3060(12GB) it only used just over 6GB with 2 workers.?
I can refit the 2060(6GB) for more tests once I have got Allan’s image working with the 5060Ti.
Any ideas for fixing the DCP Curve problem.
Regards
Bryan
PS:_ I will return to the recommended drivers and retest when I have time.
@Allan A tricky one but I can’t make sense of the results with your image!?
I cloned it 39 times (Ctrl J) , the fastest way to make a batch, of 40 images in this case and my results with the 5060Ti(12GB) suggest that it is “impossible” for you not to be having problems!?
My first test was after a test of mine so we have 8630MB with 3 workers and a time of 1minute.
I downloaded both images with the corresponding dop file, copied everything into my PL9 folder, started the app, downloaded and applied a new profile, but the first image (the portrait of the girl) opened and disappeared again with an error message.
To fix the missing DCP profile, I tried applying the regular camera profile as usual, but that didn’t work.
Then I checked Allan’s DOP file and found that he had used a DCP profile from LR ( which gets referenced in PL with this absolute path ).
( BUT only afterwards did I notice the different camera model ! )
Since I’m still using the old LR 5.7, I downloaded and installed the latest Adobe DNG converter and applied a profile…
I restarted the PL9 and a new camera profile ( E-M1 Mark III ! ) appeared. And bingo – PL opened the first image.
.
The reason:
Allan had applied the profile for the E-M5 Mark II ( his second image ! ), perhaps to make them look identical … and when I opened both images in PL, the right profile for the first image was “not recognized.”
Exporting both images as TIFF was no problem (NVidia RTX 4070 12 GB, driver 577.00) and, as shown, the first image with many AI masks took significantly longer than the second.
@Allan The images aren’t actually much smaller! It might be the drivers but I think it is the fact that you are using GTX cards with their drivers and I am using RTX cards with their drivers.
I re-fitted the 2060(6GB) and it was a disaster! With the drivers I had been using with the 5060Ti, PL9 dumped at the start of each and every export that I attempted!?
I found an alternative driver and loaded it and attempted an export and got this
i,e, it is taking PL9 forever to export each file.
So I will remove the 2060(6GB) from the machine and retire it gracefully because it currently appears to be useless when AI is mixed with NR exporting using either DP3 or XD2s.
You have been lucky but I understand your concern and others are currently struggling with PL9 and hoping for a fix.
@Wolfgang Thank you for your efforts. I am now truly tired of installing new versions of NVidia drivers and I have had more than enough fitting and removing GPUs to last me for a long time…
The only explanation I can suggest is the one at the start of this post i.e. GTX cards and GTX drivers because XD2s swamps my 6GB card and DP3, which @allan used successfully is even worse!?
The reason: Allan had applied the profile for the E-M5 Mark II ( his second image ! ), perhaps to make them look identical … and when I opened both images in PL, the right profile for the first image was “not recognized.”
Sorry, I forgot about that. I have 4 different Olympus cameras that are much about the same output wise. So, I created my preset using a Lightroom DCP and I apply it to all of the camera models.
At first, I didn’t understand why PL couldn’t open the girl’s portrait properly, or rather, why I couldn’t replace the missing DCP file (in PL’s dop-file saved with & referring to an absolute path) with a standard camera profile, which is my usual approach to solving this problem.
As it turns out, PL somehow didn’t correctly detect the E-M1 Mark III camera and only suggested the camera-lens profile for your E-M5 Mark II. But then it did, I applied the correct camera-lens profile, and everything was fine.
One observation that I made was with the “Maximum cache size”. On both computers, the default is 5000Mb.
I read that this should be half of the ram size, so I set it to 16000Mb. While there were no crashes, the computer seemed to run “rough” like a car with a rough idle. The “Full preview in progress” was slower.
Returning it back to 5000Mb fixed this.
Any ideas on this?
(The number of simultaneous processed images is 2)
It’s the same for me – and I’ve set the cache to 20,000 MB. It’s on a separate NVMe drive that I use as a scratch disk for various things. But I have no better idea what the “right” cache size is.
@Allan I have been testing 9.02 this morning and got some good and some bad results with
My 5900X is happy to be put to sleep and has a large illuminated power switch which flashes when asleep and which immediately powers the machine up when pressed and, more to the point, when woken after sleep the GPU memory has reverted to a low starting point.
So here is the results of runs with 40 copies ([M]aster and 39 VCs) of the second image you supplied.
After various simple adjustments have been made (changing DP3 to XD2s for example)
The memory used is not the highest I have seen while testing with the 5060TI(16GB) but is typical at 8859MB
Please note that any changes to the image options, no direct edits but just selecting all images or changing XD2s to DP3 or vice versa etc. were done before PL902 went to “Sleep” (along with the rest of the machine, i.e. 4899MB
@Required My tests with both the GPU benchmark Nikon images with AI selections crudely applied by me and tested as physical images and as one physical image and 39V Cs and Alan’s second image have varied wildly!
Using my 5060TI with 16GB the tests have reached as high as 12778 GB and in none of my tests today have I made a single edit change except to select all images or change the NR. Whatever DxO are doing in PL9 the GPU memory management is all over the place, there are other non-technical terms I could use but I want to be able to go on posting.
My recommendation for those with limited GPU memory is to try using “Sleep” to purge the GPU before an export providing
Set up PL5 ready for export so that all that is needed is the actual export command.
Your machine can reliably be put to sleep and restored
Your machine can be quickly put to sleep and then restored from sleep, I am lucky with the 5900X that the keyboard has a dedicated sleep key and it is conspicuously obvious that the machine is asleep.
PS:- This isn’t going to help with editing issues but the occasional purge of GPU memory might not be a bad thing, unless it takes forever or is not reliable!
