I am at a point where I do not believe I could outsource the net code for the kinds of games that I'd like to develop anymore. By extension, this sort of ruins my ability to use any commercial product like UE5/Unity/et. al.
Networking is very fundamental to some types of games and the expected player experience. Grabbing a generic "this is multiplayer :D" box off the shelf and cramming it into your concept will likely fail miserably if you are seeking a high-quality outcome.
Certain flavors of net code cut extremely deeply into the game engine concerns. Rollback net code is a pretty big dragon in that it suggests your simulation must remain deterministic at all times. Ideally, if you are planning time traveling style net code, you would first have this problem cooking before you send a single byte across the wire.
It's interesting in that respect what the developers of Frost Giant Studios, comprised of ex-Blizzard RTS devs, are planning for Stormgate. Complex RTS games need a deterministic ("lockstep") simulation to be viable over the network. On the other hand, they are using UE5, which as you indicate, has a completely different server/client model.
What I've gleaned from interviews is that the simulation part of the game including the netcode will be completely custom. The way I understand it, they're effectively only using Unreal Engine for rendering, sound, UI, cutscenes and so on, but all the gameplay logic is their code. Customizing an off-the-shelf engine to that level has to be highly nontrivial, and probably out of reach for smaller projects.
Actually as long as you have access to the lower level APIs changing that stuff out whilst not trivial isn’t that crazy. The big win with UE in particular is then phenomenal art pipeline and tooling which would take an age to reproduce in a custom engine.
> The big win with UE in particular is then phenomenal art pipeline
This is the only fear I have with 100% DIY - If I want to go much beyond 90s graphics, I will need to do a lot of extra work (aka "draw the rest of the owl") to even begin talking to artists.
That said, there are ways to build games that do not require elaborate art tool chains. AAA visuals are the antithesis of my target. I am perfectly happy trying to do art in code until the wheels pop off the bus and I have to retool for an external art team.
Yeah, it's a shame because multiplayer was an afterthought for the current major engines that you listed. Off the shelf multiplayer can work really well if the engine was designed for multiplayer. The Source engine always comes to mind as an engine that has proved itself with many high quality multiplayer titles. I wish Valve made licensing Source easier/cheaper.
>Grabbing a generic "this is multiplayer :D" box off the shelf and cramming it into your concept will likely fail miserably if you are seeking a high-quality outcome.
How many completed games have you shipped with your custom netcode and engine?
Unity multiplayer implementations have been very awful indeed. But UE4 netcode was pretty decent for local small room (under 16 people, whole scene loaded at once) scenarios out of the box.
Networking is very fundamental to some types of games and the expected player experience. Grabbing a generic "this is multiplayer :D" box off the shelf and cramming it into your concept will likely fail miserably if you are seeking a high-quality outcome.
Certain flavors of net code cut extremely deeply into the game engine concerns. Rollback net code is a pretty big dragon in that it suggests your simulation must remain deterministic at all times. Ideally, if you are planning time traveling style net code, you would first have this problem cooking before you send a single byte across the wire.