AMD Radeon GPU Profiler vs GFXReconstruct

AMD Radeon GPU Profiler (RGP) is a free low-level optimization tool for examining workloads on AMD Radeon GPUs. AMD positions it for PC game developers and developers optimizing applications for AMD RDNA hardware. It analyzes graphics and asynchronous compute usage, event timing, pipeline stalls, barriers, and bottlenecks. Frame and queue views show command-buffer submissions and interactions among graphics, compute, and copy workloads. Wavefront analysis links wavefronts to launching GPU events and displays cache information, while instruction timing reports average latency for AMD RDNA instructions to help locate shader hotspots. RGP supports DirectX 12 and Vulkan graphics, plus OpenCL and HIP compute. It runs on Windows 11 and Ubuntu 24.04 LTS, with Linux limited to Vulkan. Version 2.7 is dated June 2026. Older APIs such as DirectX 11 and OpenGL are unsupported.

GFXReconstruct is a free toolkit for capturing and replaying graphics API activity, including Vulkan and Direct3D 12. Capture records API calls, parameters, and state in a file; replay can run without the application that created it. The repository includes capture layers and libraries, replay, capture information, compression, SPIR-V extraction, conversion, and optimization tools. Listed uses include reproducing defects, examining API usage, profiling performance, regression testing, and platform bring-up. Vulkan capture and replay are listed for Windows, Linux, macOS, and Android, while Direct3D 12 capture and replay are listed for Windows only. The maker names Samsung Sokatoa, Arm Frame Advisor, Google Android Performance Analyzer, and Qualcomm Snapdragon Profiler as tools powered by GFXReconstruct. Vulkan users can also use it alongside Vulkan validation layers and the Vulkan Configurator. OpenXR support is experimental, covers limited features and a small number of apps, and is not enabled in the SDK-built version. Cross-operating-system replay may work in some cases, but is not generally supported unless specified. The project uses the MIT License.