Keep in mind that not all performance issues will show up on such graphs and not all can be found using profilers. While with some experience you'll be able to see that you could benefit from setting affinity to your threads or changing which threads execute on specific NUMA nodes, it might not always be that obvious to see that you've forgotten to disable power-saving features or would benefit from enabling or disabling hyper-threading. Information about the hardware you're running on is useful, too. Sometimes you might see the SIMD registers of your CPU being used, but the code still doesn't run at its full speed: you might be using SSE instructions instead of AVX ones, AVX instead of AVX2, or AVX2 instead of AVX512. Knowing what specific instructions your CPU is capable of running can be golden when you analyze the profiling results.
Solving performance issues also requires a bit of experience. On the other hand, sometimes experience can lead...
