The stock advice to new game developers is “make a game, not an engine.” Elias Farhan argues that advice has hardened into dogma, and the market still rewards custom tech: about 13% of games released on Steam in 2024 used a custom engine, down from 71% in 2012, but those same games accounted for 43% of units sold.

The post opens a series on porting Soup Raiders off Unity onto a native C++ engine targeting 60 Hz on Switch. Before the technology, he lays out the case for owning the layer between gameplay code and the OS.

Reasons that hold up:

  • Learning that transfers. Hand-implementing cascaded shadow maps and occlusion culling in OpenGL means you read the equivalent Unity settings correctly instead of guessing — and can confidently switch culling off when profiling shows it earns nothing.
  • Three kinds of independence (after Maxim Kiselev): technical, where a blocking engine bug means waiting on the vendor; legal, where licences get revoked, as with Improbable/SpatialOS in 2019 and a student the day before a demo launch in 2026; and financial, where Unity Pro costs US$2,310 per seat per year past US$200k revenue while Godot is MIT and free.
  • Control over the build, not just the frame rate. Nitrous Engine compiles in under 30 seconds; on the same machine a Unity IL2CPP build took 1.5 hours where UE5 source took 40 minutes. Farhan kept a second computer for nightly multiplatform builds of Beach Slap.
  • Specialisation by design. Noel Berry’s engine reads Aseprite files directly and ships per-game level editors; Animal Well’s engine pairs rim lighting with a full-screen Navier–Stokes fluid simulation. General-purpose engine features are skipped on purpose.
  • Mechanics that would not exist otherwise. Noita simulates every pixel, Miegakure needed 4D collision and lighting built from scratch, Teardown grew out of voxel destruction tech.

Where commercial engines still win: multiplatform. Unity console ports are close to magic compared with hand-rolled ones, though custom engines keep a clear edge on Playdate, Arduboy, PS1, and Game Boy.

He is careful with the costs. Tyler Glaiel’s verdict is that you “most likely won’t” save money, and Nikita Lisitsa’s is simply “it is hard and time-consuming.” The unbreakable rule that survives all of it: make a game at the same time as the engine, because an engine is nothing without a game.

The framing worth keeping is narrower than “roll your own everything.” Build a custom engine for one specific game and its platform constraints, not as a general-purpose Unity replacement — the case rests on specialisation and on owning the parts your product actually depends on.