ncu-cuda-profiling

Automating CUDA kernel profiling involves running Nsight Compute to collect all metrics and saving the reports for persistent storage. This Skill structures the output into a clear report and CSV data directory for reproducible workflows.

Nsight Compute profiling identifies bottlenecks by generating AI-ready analysis summaries of collected metrics. It highlights specific performance limiters such as DRAM, L1, SM, and occupancy issues to guide kernel optimization.

Yes, comparing CUDA kernel profiling results across optimization runs is supported by storing Nsight Compute reports and CSV data persistently in a structured directory, ensuring reproducible workflows for later measurement.

The best way to get actionable optimization suggestions is by collecting full Nsight Compute metrics. This process produces AI-ready analysis summaries that target bottlenecks like DRAM and occupancy, providing concrete guidance for improvements.

No, you do not need to manually configure metrics for diverse CUDA workloads. This Skill automates full metrics collection with a single command, applying Nsight Compute profiling across various kernels and saving the results locally.