OK. Thank you for that.
So I am therefore working on adding a remote ML-server to my Immich-server. The Immich-server itself is on a Synology DS923+, with no GPU capability.
So, on the same LAN i am adding a ML-server running on a Windows 11 Pro 24h2 pc, in which I have put an Asus graphics card with a Nvidia TX1070. As far as I have read, this however, must run Docker on Linux via WSL2, since a VM on a VMWare Workstation would not be able to load the necessary Nvidia drivers.
The both work with 2,5 Gbps on this LAN.
I have not yet finished to get that arrangement to work flawlessly.
But all that in order to have the ML run in an acceptable speed. There are 85k photos, and they were processed with 1.000 per hour, i.e. 85 hours per user added, if we just not all share one and the same admin user (with unpleasant security implications).
I am very fond of Immich and the really brave work I see unfold here from all contributors.
I got a much newer version on my smart phone a couple of days ago, and I look forward to have for instance the map features to the pc-version as well.