Working under Windows 10. During editing an image in Darkroom, Darktable just stopped and dropped out. This happened only on a certain image edit. If I discard the editing history, everything works fine.
The log file contains a huge amount of lines for the time, but from these lines I can not find out what went wrong.
I would like to know the cause of this unusual behaviour, to be able to avoid it in the future. I hope there is someone out there that can help …
Do you use OpenCL? If yes, what card? Can you reproduce the crash? If yes, can you provide a file and instructions?
Thank you for a very fast response.
Yes I’m using OpenCL (activated in preferences/processing). The drivers are up to date.
Yes, I can reproduce the crash using the image raw file and XMP file:
Pericnik_231002_123644_J_3484_R00.pef (49.7 MB)
Pericnik_231002_123644_J_3484_R00.pef.xmp (14.3 KB)
I can also provide an excerpt from the log file if that can be of any help.
Please do. The last lines might hint at the cause.
Tuning for memory transfer (I think that’s the same as pinned memory) is mostly for old ATI/AMD cards, and will probably hurt performance. I don’t think it should cause a crash, though.
Do you also get a crash without OpenCL?
1 Like
I agree. I actually did a bit of testing when these preferences were introduced and for me and my Nvidia 3060Ti… the best option across a number of test image was no tuning… FWIW
I had this bug a few times where DT crash only when loading a specific image, and this was related to editing history as well. My guess is that it is related with masks copied from a picture and pasted onto the buggy image. When I redo the history from scratch the bug disappeared.
I played a little bit to investigate the crash/no crash situation at different OpenCL settings with the following results:
What is your darktable resources set to? Since tuning for memory size seems to help avoid the cache, maybe you are running out of memory?
Machine memory (32GB installed) does not seem to be a problem - according to Windows task manager the occupied memory is rarely over 40% during my image processing sessions. Anyway, here are my settings in darktable:
However, I have not been tracking the GPU memory (4GB installed) - for a modern GPU this is a bit low …
I meant GPU memory. The resources setting influences GPU memory usage, too (read more in the docs).
https://darktable-org.github.io/dtdocs/en/special-topics/mem-performance/#darktable-resources
Yes, the restricted GPU memory could be causing the problem. However, I made a quick check with a few potentially problematic edits, and the GPU memory usage remained always below 3GB (typically peaked at 2,8GB which roughly corresponds to the formula given in the user manual for a 4GB GPU card) - that is with my current tune setting “memory size and transfer”.
With my previous setting, there could have been a memory problem.
Anyway, currently my darktable installation is running without problems. In the future I will probably have to consider replacing the GPU card.
Thank you for your help in this discussion.
What image edit? what module? Is it a mask feathering?
I looked thru the 360k rows of log file but I dont see a clear issue.
The raw image and pertaining XMP file are linked above (Oct 4 @ 17:00)
Not directly connected with the crash (that happens during editing) but maybe also worth noting, is that during export the GPU memory is completely full (~4GB according to the Windows Task Manager) and on some edits switches the processing over to the CPU in mid export process.
After a few such exports (multi file export job) sometimes I get a message stating there is a problem with the OpenCL configuration, and after that the remaining exported files are broken. In such a case restarting darktable removes the problem …
I looked at your IMG and xmp. I don’t get a crash.
Right now I don’t understand the steps to recreate your issue. That’s why I asked what module is the one you are using when the crash happens.
The crash (actually darktable doesn’t freeze, it just drops out of memory) happens sometimes as soon as the thumbnail is refreshed in lighttable, or when I enter darkroom, just before the main image is displayed. So, I do not know, which particular module causes the crash. Maybe a good guess could be the latest (topmost) in history. This is probably the one I was working on, when the first crash occured …
Did you look at the image in a Windows installation?
I used Fedora 38 KDE current master.
The log should have some info on the crashes. I did not see it. If you want you can try the current development build installed in a test folder.
There may be differences in Windows and Linux installation, especially in memory handling.
If I use a development installation (installed in a separate test folder) is it advisable/possible for this version to access the same database&images as the regular stable installation?