Real World Comparison, GC vs. Manual Memory Management

I've had very similar concerns about GC vs. Manual memory management for games. It's good to finally see some sort of quantitative research and investigation, rather than people throwing around their opinions.

It's interesting that GDC is so much faster than DMD, especially given DMD is supposed to be the reference implementation. But then he is using plenty of optimization flags with GDC.

I don't know what sort of GC D uses, but incremental or generational garbage collection would probably be better for smoothing out differences in framerate due to GC rather than running an entire GC cycle every frame.

It's actually a pretty impressive game for only 3 months; I'm sure I couldn't code a 3D game with particle effects and that sort of lighting from the ground up in 3 months.

Overall for a performance hit of 4ms per frame, It would probably be worth using a GC when you consider all the errors and issues it would eliminate from game development. For me personally it would totally be worth it developing small independent games. Maybe not so much for AAA games where performance matters.

Interesting comparison, but it should be noted that DMD uses pretty much the simplest possible GC (non-generational conservative mark-sweep).