In a world of shared GPUs, what tools do people use to profile their GPU workloads. Are there code level profilers like for CPUs? I typically hear that people d

In a world of shared GPUs, what tools do people use to profile their GPU workloads. Are there code level profilers like for CPUs? I typically hear that people don’t often have issues with the code running on the GPU itself but rather ensuring it is fed information quickly enough to stay saturated. How do people measure this and find out what to improve, especially in a shared setting?