unreal-rendering-performance

Start with profilegpu or a GPU trace to identify the dominant pass, then use subsystem visualizations like Shader Complexity, Nanite Visualization, or VSM Cached Page views to locate causal content. Change one lever at a time and re-capture the identical workload to verify the delta.

Use stat sceneRendering, stat rhi, and the Primitive Debugger to measure primitive, section, and draw-call counts. Reduce submissions by instancing repeated meshes, applying HLOD for distant groups, setting cull distances, and removing material slots only when section submission is measured.

No, Dynamic Resolution only reduces GPU pixel work by lowering screen percentage when frames exceed the GPU budget. It cannot recover a CPU or render-thread bottleneck, so you must first confirm the frame is GPU-bound with stat unitgraph and stat raw.

Yes, MegaLights is Production Ready in UE 5.8 for many dynamic shadowed local lights, with hardware ray tracing recommended. Directional Light support is opt-in and disabled by default, and quality degrades when too many important lights compete for its fixed per-pixel sample budget.

VSM spikes come from cache invalidation caused by moving lights, deforming geometry, WPO, large bounds, and rapid disocclusion after camera cuts. Use the Cached Page and Shadow Casters visualizations to find invalidators, then tighten bounds or disable distant deformation and shadow casting.

Nanite does not make materials, masked overdraw, WPO, shadows, streaming, or storage free, so measure those costs separately. Preserve conventional LOD fallback for platforms lacking Nanite support, and never use triangle count alone as a performance target.