Some time ago I reacted to a post where another user was complaining that the export worker(s) were not terminating automatically after an export was completed and continued to take VRAM and RAM, which could have been freed up.
I investigated the behaviour of the export workers and submitted a support (fault) request.
On that release, and on PL8.8/PL8.9, 1 export worker was set running for every copy determined by the highlighted setting below
The problem was what PhotoLab did (or didn’t do) after the export was finished which seemed yo vary from one setting value to the next etc.
At the time I also thought I remembered that all export copies were terminated in past releases but DxO Support told me that wasn’t the case and never had been the case!?
Running a test on my 5600G with 3060(12GB) on PL8.8 and PL8.9 this is what happened on those releases
and this happened in all cases, i.e. one export copy for each setting to a maximum of 6 (or 7 in the retest) in my testing and all copies were closed at the end of the export.
Just out of interest these are the figures for the exports of 10 images
PL8 has become a well behaved product, although some will argue that leaving the export workers running will save export time, by not having to re-initiate them for the next export. On short runs like the one that I ran for this test that may well be true.
But that behaviour has changed in PL9.1!
There appears to be only one copy of the export worker regardless of the ‘Preference’ settings, which can now reach a maximum of 6 copies and no more, at least on my PL9 version.
In the past when export worker copies stayed after an export was complete they could be terminated by the user and that would return some VRAM and some RAM. That now seems to be being done automatically on PL890 but I need to see if that actually results in any reduction in the VRAM usage.
It appears that the new PL9 export worker becomes a permanent fixture after one export has been undertaken?
It has been suggested that the export worker “times out” after 2 hours or 7200 seconds unless the user hacks a field and reduces that time.
For details see the post identified below but please be aware that no specific details of where that field is held are given. The reasons for not making that “hack” public are made by @Wlodek in that topic and repeated by me in PL9.1.0 Win with NVidia 581.42 Games drivers still fails - #78 by BHAYT.
Because I was testing the behaviour of PL8 versus PL9, I left any AI out of these tests,
I haven’t marked this as a bug because it is definitely a feature, for better or worse.
Update:- I have a potential problem with the figures and need to run the tests again to confirm them but the rest of the post is as accurate as I can make it.
