@Stenis but I am afraid you looking at the processor performance and I believe the current problems are with the somewhat haphazard way that PL9 is handling the AI modules that need to be loaded into the GPU memory for AI masking and then for AI exporting.
Testing on my 5600G, which was then equipped with a 2060(6GB) GPU, was pretty much a disaster with PL900 so the 3060(12GB) was moved to the 5600G from my 5900X and I managed to get a 5060Ti(16GB) for a good price to replace the 3060(12GB) in the 5900X.
This is a run that I just did on the 5900X with the 5060Ti(16GB) after a reboot onto an earlier NVidia driver using one of @Allan’s images.
PL902 went straight into the correct Folder and I exported just the first image and I got this
8529MB of GPU memory used but I have seen higher GPU memory usage with both PL901 and PL902, so far as high as 12778MB. Whether having a 16GB GPU simply gives more space which PL9 feels it can “consume” haphazardly I am not sure.
But if anyone is able to buy a new GPU card, i.e. laptop users have a bigger problem, then I would suggest that they do what I did and buy a card with at least 12GB and preferably 16GB and/or wait until DxO have resolved the “excessive” use of GPU memory or when they spot a GPU with a larger memory at a “good” price.
Also see my post at PL9 on older computers - #20 by BHAYT for more details of runs with PL902 earlier today.