Hi ##### on you OS (win/linux) question ##### i use both win and linux. I can't see any difference. Besides that linux saves you the 80-100€ per node that you won't have to shuffle over to Microsoft. If you buy 1-4 PCs to speed up your work that's up to 400€ "saved". Maybe some of this saving could even be donated to Carl to support his dcpomatic work ;) #### about the hardware #################### a while ago there was some discussion on this list about best buck for frame/s. Back then it was based on "old" dual-cpu xeon systems with lots of cores and two cps. You can/could(?) buy these used workstations/servers quite cheap. This was all before the big ryzen ripper time. But i think you will get used xeon E5 systems that the benchmark page in dcpomatic.com lists at about 16-20 frames/s to a price per system where you won't even get the ryzen-cpu itself ;) and based on the info from the benchmark the AMD 1950 delivers "only" about 24f/s #### about dcpomatic network setups ### The "network/cluster" function of dcpomatic is nice to have but is not maxed out design wise so far. Even with my "low end" i5/i7/i5 minicluster setup on a gigabit Network neither the network nor the two i5 client cores are "maxed-out". Therefor in my case it would be best to run three singe-none-network dcpomatic programs converting three different films parallel. With three dcpomatic_batch converts running you even get 3 waiting lines in parallel. My point here is the question "is a multi-multi-cpu-xeon/ryzen cluster workable on a gigabit network" in addition to the not so perfect cluster setup in dcpomatic itself. Maybe some other helpfull soul can give you some insight into network saturation with big core count systems. Carsten Kurz via DCPomatic writes: -- Markus Kalb Filmkreis an der TU Darmstadt