GPUs

Performance analysis in PC gaming has become more complex with the introduction of DirectX 12 and Vulkan, incorporating technologies like asynchronous compute, hardware ray tracing, and machine learning-assisted denoising. Understanding GPU performance requires advanced tools to analyze workload distribution and shader behavior, with NVIDIA Nsight Graphics being a key resource. NVIDIA Nsight Graphics is a graphics debugger and profiler for modern graphics APIs, featuring the GPU Trace Profiler, which provides detailed insights into GPU execution and performance bottlenecks. Recent updates have improved its utility, including a Flame Graph for the Shader Profiler and support for NVIDIA's DLSS in D3D12 applications. The Peak-Performance-Percentage analysis method focuses on identifying GPU workload constraints and optimizing utilization rather than relying on assumptions about performance bottlenecks. This methodology is applied using Nsight Graphics to analyze GPU traces from the game Cyberpunk 2077, contrasting traditional rasterization with path tracing enhanced by DLSS Ray Reconstruction. The test setup included an Intel Core i7-14700K CPU, 32 GB RAM, and an NVIDIA GeForce RTX 4090 GPU, capturing two scenarios: one with raster-only settings and another with path tracing and DLSS. The analysis revealed that the rasterized frame's most costly aspect was lighting processing, while the path-traced frame showed increased total frame cost and a hybrid workload involving rasterization, ray traversal, and AI-driven processes. Key findings indicated that modern GPU performance is heavily influenced by memory behavior rather than just shader arithmetic, with significant time spent on compute-driven lighting processing and data movement. The analysis also highlighted that ray tracing performance is constrained by memory efficiency, indicating that performance does not scale linearly with RT-core throughput.