Profile CUDA/GPU simulations with Nsight Systems and analyze bottlenecks statistically.
npx skills add https://github.com/hkqai/MatClaw --skill nsys-optimizer-hkqai
Please help me install this Agent Skill. Skill: nsys-optimizer Source: https://github.com/hkqai/MatClaw/tree/main/skills/nsys-optimizer Command: npx skills add https://github.com/hkqai/MatClaw --skill nsys-optimizer-hkqai
Slow or inefficient CUDA/GPU simulation code hampers research productivity and leads to excessive compute costs. This skill provides a systematic workflow to profile, diagnose, and optimize such code using NVIDIA Nsight Systems.
nsys to capture detailed kernel and API statistics for any simulation script.Ask the nsys-optimizer skill to profile and improve your CUDA simulation by providing the path to the scene script.
High-intent search queries and answers about installing and using this skill.
To optimize slow CUDA simulation code, this skill orchestrates Nsight Systems profiling to capture kernel statistics, diagnose bottlenecks, and apply statistical A/B testing for iterative speed-up verification.
GPU kernel bottlenecks are diagnosed by parsing nsys profiling data to highlight low-parallelism kernels, synchronization points, and memory-bound workloads causing the performance degradation.
Yes, you can validate CUDA optimization impact by executing repeated nsys profiles, aggregating the data, and applying Welch's t-test to statistically evaluate the performance differences between configurations.
The best way to trace hot paths in CUDA scripts is by adding NVTX annotations, which the skill uses to guide correctness testing and iterative re-profiling until performance goals are met.
Yes, scipy is required as a dependency to perform the statistical A/B analysis and apply Welch's t-test when evaluating the impact of your CUDA kernel optimizations.
The nsys-optimizer identifies synchronization points, low-parallelism kernels, and memory-bound workloads within GPU simulation workloads to systematically locate and resolve hot paths.
igamenovoerConfigure CUDA build environments for Pixi projects with CMake and nvcc.
VCERSIdentify CUDA performance bottlenecks from Nsight Systems profiling data.
yo-stevenProfile CPU, GPU, and communication timelines from .nsys-rep captures.
alibabaAnalyzes GPU kernel performance using Roofline models and profiling data.