perf-nsight-compute-analysis

To identify CUDA kernel bottlenecks from NVIDIA Nsight Compute profiling reports, the Skill analyzes key metrics like SOL%, occupancy, and cache hit rates. It automatically classifies whether your kernel is compute-bound, memory-bound, or latency-bound based on the extracted profiling data.

The best way to classify GPU performance bottlenecks for CUDA optimization is by interpreting Nsight Compute outputs. This Skill evaluates resource utilization and memory hierarchy analysis to determine if memory access patterns or compute utilization are limiting kernel performance.

To analyze Nsight Compute metrics for optimizing memory access patterns, the Skill extracts and interprets cache hit rates and memory hierarchy data. It uses these profiling outputs to recommend targeted optimization strategies for your CUDA kernels.

Yes, you need Python libraries to parse Nsight Compute profiling reports. The Skill requires these dependencies to extract instruction throughput, occupancy, and other critical GPU performance metrics for deep inspection and bottleneck detection.

Yes, you can use this Skill to determine if your CUDA kernel is compute-bound or memory-bound. It performs bottleneck classification by evaluating SOL%, instruction throughput, and resource utilization from the Nsight Compute profiling data.

During Nsight Compute performance analysis, the Skill extracts specific GPU metrics including SOL%, occupancy, cache hit rates, and instruction throughput. These metrics support deep inspection of profiling outputs for CUDA program performance tuning.