That is an option, yes, but I actually meant simply entering ‘profileGPU’ into your command line and looking at the frame time graph. It only accounts for GPU behavior, so if this is a CPU bottleneck you can’t tell from there, but it’s good at giving you a rough frame breakdown for a given moment in time. It’s my first tool to solve perf issues before I look at more nuanced (read: time-consuming) sources for data.
Good example I had: I was getting a terrible framerate on the GPU side after loading up a scene. I had no idea what could be the cause because I was testing very well the last time I ran the scene, and profileGPU showed me that some meshes I set to nanite weren’t saved as nanite, and I was eating the cost of lumen scene update and shadowing costs for non-nanite geo. Little things like that can make a world of difference.