Hi, By following the following modification: Getting GPU-Accelerated PaddlePaddle Working in Docker on QNAP NAS , my system has been running very well for several weeks. I put an unloading after 8 hours of my inference, but whether it’s the reloading or the change of inference, I no longer note any problem.
Regarding the RAM, I decided to buy 2*KSM56E46BD8KM-48HM.
Not knowing the weather about the price of RAM, I still felt like letting the system breathe a little more (dockers, VMs, etc.). We’ll see later if my purchase was wise, but for the past two weeks, I haven’t been at 60% RAM used with my 32GB of RAM by the system, but 35% to 50% depending on the load with my 96GB.
Note: I currently have RAM ARC at 50% reserved. We’ll see if I can go back a day later.
Have a good day