I thought about writing down all variables that make up our current situation. I feel that they are important to the situation, but they are not directly relevant for this question. Therefor I decided to post those considerations here: https://pastebin.com/h0rLWvsn .
The Lack of Info Online
I've been searching for sane, recent render node setups, but I keep finding threads where people cram the most powerful Xeon they can find in their nodes and include the best graphics card they can afford. That might work for an editing rig, but from a server/horizontal scaling point of view that seems ridiculous.
To my understanding, smaller, cheaper, dispensable nodes will be capable of handling more subframe at the same time. Each subframe will take longer, but all in all that should not matter that much since overall time will be cut by the multitude of cores.
The CPU vs GPU battle
The GPUs in our editing systems are indeed about 4x (at least) faster than the same renders on our editing system CPU. However, some scenes populate 10GB peak memory and graphics cards like those are just to expensive to make sense. Also putting a €1000 GPU in a €1000 machine leaves you with 1x €2000 node, whereas it could also give you 2x €1000 CPU nodes. They can be working on multiple projects, or multiple subframes, and if one breaks only half of your capacity is lost.
However, I don't see people taking this road online, so I'm wondering if my way of thinking is wrong somehow. If so, can someone please point out where my reasoning is wrong. If not, what CPU is a valid choose for this (frame/€)?
What are good resources to find more info about this, specifically for CPU rendering nodes?
