@stuck Thank you for testing the images and being so successful where I am having problems.
Having just taken my car in to have its MOT and walked back home I sat down to do some tests.
All tests were conducted on my 5600G(32GB)-3060(12GB) system with a single export stream to 100% jpgs. The 5600G has an embedded GPU which I use to drive the monitors.
I started with an XD2s test of the St Pancras image and it is essentially the same as the DOPs I made available but with XD2s set for the [M]aster and the 6 VCs and got a surprising result, for me!!
This was the starting point for the test, i.e. effectively what was in the DOPs from the previous run, 3 successful and 4 failures.
The result for the new run was, success for all images!!??
I couldn’t believe the results so I restarted PL9.5.0, cleared the Database and ran it again, successfully with essentially the same time 4 minutes and 13 seconds for 7 exports with XD2s applied creating 100% jpgs to an NVME!
BUT I then followed it with an attempt to run the NO NR test which keeps failing and it didn’t fail but ran so s l o w l y !?
So I cleared the database and restarted PL9.5.0 with the NO NR directory and got my old “favourite”
This is the ‘Vehicle’ preset that keeps failing on this test and actually failed yesterday when I did each copy one by one
So counting the [M]aster as test 0, test 2 fails (consistently)in the NO NR scenario, but passes in the XD2s scenario when doing all exports in 1 batch. Please remember that tests 5 and 6 also contain the ‘Vehicle’ preset test alongside other preset tests (‘Sky’ and ‘People’) plus an AI mask selection tests.
DxO have given a much more detailed description of potential issues as reported by @KeithRJ over at AI stability update received from DxO
PS:- The drivers used were 595.71
