Since installing build 643 on my W11 system any image I process using either Deep Prime or DP XD3 takes a very long time to process and produces a very small image file that is completely dark. Processing using Standard Output produces a good image but no noise reduction. Attached are my RAW (ORF) and output from each of the three output methods (denoted by the suffix in the name).
Please help me resolve this issue.
O4210073.ORF.dop (10.2 KB)
O4210073.ORF (17.8 MB)
No problem here with PL9.7 (build 643), RTX4070/595.79, Win11 25H2 (26200.8246), DP3 or XD3. Added some softening on the negated Flowers mask:
Glad to know the release works on someoneâs system. It doesnât on mine. Mine worked fine with the prior release. Is there a way to revert?
I found the DopCor.txt files - Iâve no idea how to use them to figure out whatâs going wrong.
DopCor.txt (24.1 KB)
Try using version 9.6.1 or lower, but 9.6.1 worked stably for me. The key is the GPU; I had a GTX 1660 Super and the same issues with v9.7 as you.
Maybe you should experiment with âWindows ML Accelerationâ if you have some choices available there.
BTW, I have exported successfully also using âJPEG for tablet or HDTVâ, like in your case, with all metadata. And we both use 4K monitor @60Hz.
The key difference seems to be in WinML Execution Provider used. Some data from my DopCor with âWindows ML Accelerationâ set to âMaximum performanceâ:
Available Devices for ONNX Runtime:
â Intel(R) Core⢠i7-14700KF [via CPUExecutionProvider v1.23.5]
â NVIDIA GeForce RTX 4070 [via DmlExecutionProvider v1.23.5]
â Intel(R) Core⢠i7-14700KF [via OpenVINOExecutionProvider v1.3.0+b130ce1]
â Intel(R) Core⢠i7-14700KF [via OpenVINOExecutionProvider.AUTO v1.3.0+b130ce1]
â NVIDIA GeForce RTX 4070 [via NvTensorRTRTXExecutionProvider v1.23.2]
Uses TensorRTRTX Execution Provider
In your log:
Available Devices for ONNX Runtime:
â Intel(R) Core⢠i7-9700 CPU @ 3.00GHz [via CPUExecutionProvider v1.23.5]
â NVIDIA GeForce GTX 1660 Ti [via DmlExecutionProvider v1.23.5]
â NVIDIA GeForce GTX 1660 Ti [via CUDAExecutionProvider v?]
Uses CUDA Execution Provider
Not sure if the question mark in âCUDAExecutionProvider v?â is OK (some errors may be âgood errorsâ, i.e. expected).
No clues in your DopCor log for dark frame export.
Your NVIDIA driver version is 591.55, which looks like some specific version provided by PC manufacturer. I doesnât seem to be a problem here, but who knows?
Iâm not sure what choices you have in your PL9 Preferences. In my case for âWindows ML Accelerationâ is set to âMaximum performance (recommended)â but I can choose also âCompatibility modeâ. I tried to set it to âCompatibility modeâ, restarted PhotoLab. The export was successful (i.e. expected image was there) and DopCor contained:
Available Devices for ONNX Runtime:
â Intel(R) Core⢠i7-14700KF [via CPUExecutionProvider v1.23.5]
â NVIDIA GeForce RTX 4070 [via DmlExecutionProvider v1.23.5]
â Intel(R) Core⢠i7-14700KF [via OpenVINOExecutionProvider v1.23.2]
Uses DirectML Execution Provider
You may try it, if itâs available for you (requires PL restart). In any case I would escalate to DxO support. There seems to be some problem with NVIDIA Cuda EP.
Thanks. My preferences were set to âMaximum Performanceâ. I set it to âCompatibility modeâ, Restarted my computer, and tried all three denoising settings. All three errored out. Then I set it back to âMaximum Performanceâ and again tried all three denoising settings. All three worked. Tried all three on several other photos. The problem is gone!
Donât understand why - and I guess I donât care - as long as it keeps working. Thanks for prompting me to try this.
The problem came right back. Iâve now reported it to DxO Support. Will update this thread when I hear from them.
DxO support had me fully uninstall then reinstall PL9.7 It now works correctly. Issue resolved!
Is the CUDA version shown now in DopCor log? My guess is that for PL9.7 it should be v1.23.2. Maybe itâs the problem fingerprint?
Good to hear it works now.
I donât see any mention of CUDA. Hereâs the DopCor.txt
Looks like it uses Microsoft DML - not as fast as CUDA - not sure how or if I can/should change it.
DopCor.txt (23.4 KB)
Strange indeed. Previously you had
Available Devices for ONNX Runtime:
- Intel(R) Core⢠i7-9700 CPU @ 3.00GHz [via CPUExecutionProvider v1.23.5]
- NVIDIA GeForce GTX 1660 Ti [via DmlExecutionProvider v1.23.5]
- NVIDIA GeForce GTX 1660 Ti [via CUDAExecutionProvider v?]
Uses CUDA Execution Provider
and after PL reinstall thereâs no CUDA but OpenVINO appears (probably too slow to use for DeepPRIME or AI masks):
Available Devices for ONNX Runtime:
- Intel(R) Core⢠i7-9700 CPU @ 3.00GHz [via CPUExecutionProvider v1.23.5]
- NVIDIA GeForce GTX 1660 Ti [via DmlExecutionProvider v1.23.5]
- Intel(R) Core⢠i7-9700 CPU @ 3.00GHz [via OpenVINOExecutionProvider v1.23.2]
Uses DirectML Execution Provider
In my case âNvTensorRTRTXExecutionProvider v1.23.2â is used (for RTX4070), which is faster and probably more stable than âDmlExecutionProvider v1.23.5â. The DirectML EP may still suffer from memory leaks like it had before. NVIDIA(/Microsoft) offers RT RTX EP only for RTX cards, so itâs not available for your GTX 1660 Ti. I donât know anything about the CUDA Provider. Looks like your previous installation was really corrupted.
I had Copilot look at it and it told me DxO doesnât offer CUDA support on the 1660Ti. It suggested I could upgrade to a 4060 card and dramatically improve performance since itâs a faster card and could use CUDA. Iâm not interested in putting more money into this old system. Not sure Iâll stay with Windows, my MacBook Air M1 outperforms this system - if I upgrade, Iâll probably buy a MacBook Pro M5 Pro. Iâm just glad this is working for now. No idea how my prior install got the CUDA stuff in there.
