Enable this option if DO PhotoLab becomes slow or unresponsive when using Al-powered features.
Anyone noticed any improvement with this on?
None for me. Using a MacBook M5 with 48GB.
Have left it off.
There might be some specific cases where this works. I’ve turned it on and off, and I don’t see any difference. I’m using Windows 11 on a fairly powerful computer, so maybe it has some measurable effect on weaker hardware? It’s described a bit here (in the “Optimize AI performance” section): Overview – PhotoLab
It says there that “This reduces memory consumption, at the cost of having to reload the models the next time they are used.” With my graphics card (AMD), this option doesn’t seem to do anything. Once memory is allocated while using the AI feature, it isn’t actually released (there are slight fluctuations, but they’re independent of whether this option is checked) until PhotLab is closed. It’s possible that this works better with nVidia graphics cards. The situation may be different on Apple computers. Perhaps we need to wait for future updates to the program? Version 10 is very “new” and may need some fine-tuning… 
By the way… considering your computer’s performance, you don’t need this option. 
What version of AMD GPU do you have? I have an RX 5500 XT-based one with 8GB dedicated GPU RAM and 16GB shared and I find it OK for generating AI masks but almost impossible to export the resulting image. Trying to avoid spending many hundreds of pounds to replace what is otherwise a perfectly decent GPU.
I’m testing two different configurations. One of them is a computer with an AMD RX6600 / 8GB graphics card, the latest version of Windows 11, and the latest drivers, which meets the requirements for AI operations. And on this computer, I’m experiencing issues similar to the ones you described. I’ve already contacted DxO technical support. They say that with the RX6600 and 8GB, it should work fine. The matter is still being investigated; I’ve received instructions on what to check. I’ll post here if I find out anything more.
Few things: try ‘compatibility mode’. OpenCL need to be enabled. Don’t use (enable) DeeprimeRendering, don’t use Loupe. Set parallel export to 1. May helps. And close all other applications (web browsers, etc.)
Thanks for the suggestion. Unfortunately I am already using compatibility mode and OpenCL. Also I have this problem when exporting just one image, not a batch.