A detailed performance analysis examining how vertex shader exports impact GPU rendering performance. Through controlled experiments on RTX 3080, the study shows that increasing vertex shader exports from 1 to 10 float4 attributes nearly triples drawcall costs due to bottlenecks in GPU memory allocations (ISBE, TRAM) and the Primitive Engine. The research reveals that unused vertex exports are optimized away by NVIDIA GPUs, float exports scale better than float4, and different GPU architectures handle vertex data differently.
Sort: