Same here (Win11).
Export times with XD2s are ±5% the same for PL9.0.2 and PL8.8, no visible differences in the output.
After intensive AI mask usage export times increased by about 4 times, probably due to GPU memory management. Could not reproduce it though, needs further tests. It might take several months to optimize GPU memory usage, but the first step looks promising. I wouldn’t switch from learning to production mode yet, though, and stay with PL8 until at least PL9.1.
To prevent constant ‘Full preview in progress’ while working with AI masks, move the cross-hair cursor out of preview window or just deactivate the AI mask button, when no longer needed.
My config:
- ‘Enable high quality preview’ and ‘Enable DeepPRIME rendering’ ON
- Z8 (45mpx), 14-bit raws loselessly compressed, used ISO 64-12800
- Windows 11 pro 24H2 with latest patches, no swap “tweaks”
- GeForce RTX 4070 with 12 GB VRAM, liquid cooled, max TDP 200W
- NVIDIA 576.80 Studio driver (June 2025)
- Single 4K monitor at 60Hz connected via DP, no HDR, no G-Sync, 8-bit mode
- Normal power management mode for GPU
- i7-14700KF, no builtin graphic unit, so about 1GB of GPU VRAM is taken away by NVIDIA/MS.
- 32GB RAM
- 2TB+4TB NVMe Kingston disks, second one dedicated to photos
- 850W Gold PSU
No overheating or power supply problems.
BTW, with PL9.0.1 trial I got “Internal error on execution stage” within first minute of AI mask test. It didn’t happen yet with PL9.0.2, after 3 hour test.
UPDATE:
I did some complex AI mask editing on a Virtual Copy with no problems, then exited PL while the VC was selected and started PL again. Got the Internal Error upon startup with this in the PhotoLab log:
2025-09-18 13:58:00.755 | DxO.PhotoLab - 1340 - 52 | DxOCorrectionEngine - Error |
Exception caught while calling ExecuteTileImage : D:\DATA\Bamboo\Builds\FWK-ONNXRUNTIMEPKG0-ORTPKG\ort_src\onnxruntime\core\providers\dml\DmlExecutionProvider\src\DmlCommandRecorder.cpp(371)\onnxruntime.dll!00007FFCAECEE0FD: (caller: 00007FFCAEC58956)
Exception(1) tid(e38) 887A0006 The GPU will not respond to more commands,
most likely because of an invalid command passed by the calling application.
2025-09-18 13:58:00.755 | DxO.PhotoLab - 1340 - 52 | DxOCorrectionEngine - Error |
InternalError: In Block SemanticMasksObjectDetection@node_71[1].Execute
In Block DxO::Node::Execute [ identifier=SemanticMasksObjectDetection m_internalMode=PreviewOneShot m_outputScale=1 m_inputROIs[DxO::EngineDef::kMainImage].StringRepr()=[ (x=0, y=0, w=4128, h=2752) ] outputs.imagesWithROIs.at(EngineDef::kMainImage)->value=outputImage=computeOutputInfoOk=true outputImageValidROI=[ ] outputROIs=validateOutputPolicy=FullOutputData outputROI=[ (x=0, y=0, w=4128, h=2752) ] invalidROI=[ (x=0, y=0, w=4128, h=2752) ] validROI=[ ] priority=-50 ] [ ctx=0000022C9D043B70 ]
Correction failed on the Execute stage
plus more errors, probably results of the above. GPU VRAM used was about 7GB out of 12GB. After deleting the VC in question, restarted PL with no problems. Looks like not only GPU memory management is still a problem with PL9.0.2 b68/Win, with AI masks used.
UPDATE2:
Abut 3 seconds before PhotoLab logged and error, NVIDIA driver logged Event ID 153 Error with source = nvlddmkm.
, but I do not congatulate the DXO team with the PL9, until it is fixed 
. Since you have some insight, what do you think would be the best Windows PC for PL9 and photo editing in general, if you do not mind me asking? 
Maybe there is another post for this, but I am trying to put together a new Windows AMD PC mainly for editing. AMD 9 9950x3d most likely. 