cuda

Profiles and debugs CUDA kernels using nsys, ncu, cuobjdump, and cuda-gdb.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/rdromer2/mambavision_tf_kernel --skill cuda
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cuda
Source: https://github.com/rdromer2/mambavision_tf_kernel/tree/main/.agents/skills/cuda
Command: npx skills add https://github.com/rdromer2/mambavision_tf_kernel --skill cuda

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Streamlines CUDA kernel development, debugging, and performance profiling.

Core Features & Use Cases

  • Comprehensive debugging workflows for CUDA kernels, including printf-based tracing, compute-sanitizer, and cuda-gdb usage.
  • Profiling and optimization workflows using Nsight Systems (nsys) and Nsight Compute (ncu), plus binary inspection with cuobjdump.
  • Real-world use cases spanning kernel optimization, memory access patterns, and performance verification with baseline timing.

Quick Start

Run a baseline profiling session on your CUDA kernel to identify hotspots and begin iterative optimization.

Frequently Asked Questions about cuda

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I profile a CUDA kernel to identify performance hotspots?

To profile a CUDA kernel, establish baseline timing and run a profiling session using Nsight Systems (nsys) to identify hotspots, then apply Nsight Compute (ncu) for targeted kernel analysis and iterative optimization.

What is the best way to debug CUDA kernels using cuda-gdb and compute-sanitizer?

Debugging CUDA kernels involves using cuda-gdb for interactive debugging and compute-sanitizer for memory error checks, supplemented by printf-based tracing to isolate logic errors and verify memory access patterns.

Can I inspect CUDA binary files to analyze kernel memory access patterns?

Yes, you can inspect CUDA binaries using cuobjdump to analyze kernel memory access patterns and verify optimizations, which is essential for structured profiling and improving kernel performance.

How does nsys compare to ncu for GPU kernel optimization workflows?

Nsight Systems (nsys) provides system-wide profiling to locate hotspots, whereas Nsight Compute (ncu) performs targeted kernel analysis to optimize memory access and compute patterns for measurable improvements.

Why does my CUDA kernel optimization lack measurable performance improvements?

Your CUDA kernel optimization lacks measurable improvements if you skip establishing a baseline timing or fail to follow a structured profiling loop with targeted analysis using nsys and ncu.