Scripting Engine Overhead: C++ Native vs Lua/C# Garbage Collection Spikes: Competitive Benchmark Analysis and Залізо Interactions

[Ella Bell]

I’ve been benchmarking scripting engine overhead across our tournament rig and the GC pause spikes are wrecking frametime consistency more than raw FPS ever does. In our title, the Lua bridge to native C++ is showing micro-stutters during ability stacking—specifically those 8-15ms hitches that show up on the frametime graph but never register as a full frame drop in the average. For anyone who’s profiled interop invocation costs, how do you isolate the marshal overhead from actual GC collection pauses when both hit the same window? I’m running a Ryzen 9 7950X3D paired with DDR5 at 6000MT/s CL30, and I want to know how much of this is CPU-bound versus how the L3 cache on the X3D chips is absorbing or amplifying these stalls.

On the C# side, I’ve been testing manual memory pool strategies versus letting the runtime handle allocation. The pooled approach smoothed out the tail latency noticeably, but I’m seeing a different spike pattern around 600fps caps where the scheduler seems to starve the scripting thread. For those who’ve optimized competitive titles, what frametime variance threshold do you consider acceptable for a tournament-grade setup, and how do you account for hardware scaling when comparing single-core scripting performance across the AM5 and Intel 14th-gen platforms?

"gl & hf in your games."

[Author]

Posts