profiler-ncu SKILL.md skillA

profiler-ncu from tongminglaic/ako4x is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.

Install it, pinned
mdr add tongminglaic/ako4x/profiler-ncu@git:20260531.0fd4b5f

Pins git:20260531.0fd4b5f in mdr.lock. Pin the fingerprint instead.

Audit grade
A, 17 of 17 checks passed
Kind
SKILL.md skill
Repository
tongminglaic/ako4x
Path
templates/skills/profiler-ncu/SKILL.md
Versions recorded
1
Latest version
git:20260531.0fd4b5f, committed May 31, 2026
Size
2,177 bytes
Repository stars
84
License
MIT

Run NVIDIA Nsight Compute (NCU) per-kernel profiling for register pressure, occupancy, stall-reason, memory-throughput, and IPC analysis. Use BEFORE architecting an optimization fix — not only after — when a hypothesis about microarchitectural behavior needs verification. NCU `Duration` is NOT comparable to bench timing; use NCU for ratios only.

Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.

What the file says

# Profiler (NCU)

Wrapper around `ncu`. Command entry: `bash scripts/profile.sh`.

Detailed reference: `ncu.md`. Top-level commands:

```bash
bash scripts/profile.sh --list
bash scripts/profile.sh --index 5                          # default --set detailed --page details
bash scripts/profile.sh --index 5 --set full
bash scripts/profile.sh --index 5 --kernel-name ".*name.*"
bash scripts/profile.sh --index 5 --sections LaunchStats,Occupancy
bash scripts/profile.sh --index 5 --env NO_GRAPH=1         # modal backend; on local, prefix the run with NO_GRAPH=1
```

## Key constraints

- **NCU under graph capture is largely blind** — graph-launched kernels go through `cuGraphLaunch` and the `--kernel-name` filter often returns "No kernels were profiled". If your kernel uses `torch.cuda.CUDAGraph`, install a module-level `_NO_GRAPH = bool(os.environ.get("NO_GRAPH"))` gate AND set `NO_GRAPH=1` — modal backend via `--env NO_GRAPH=1`, local backend by exporting it in the shell (`run_local_profile.py` has no `--env` flag). See `ncu.md` "CUDA graph capture interaction".
…

Read the whole file at its exact version.

How to install

The exact contents, by fingerprint
mdr add tongminglaic/ako4x/profiler-ncu@sha256:671d0c65b298e0e6

The line at the top of this page pins the label git:20260531.0fd4b5f, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.

Badge

mdr badge

[![mdr](https://markdownregistry.com/badge/art_dghrhj3h4zfsfwyg.svg)](https://markdownregistry.com/a/art_dghrhj3h4zfsfwyg)

3 badge image fetches in 30 days. What the badge shows.

Versions

The registry has recorded 1 version of this file, committed on May 31, 2026.

versioncommittedcommitsizeaudit
git:20260531.0fd4b5f latest2026-05-31 0fd4b5f 2,177 BA view

Audit of the latest version

A  17 of 17 checks passed. Fixed automatic checks with no AI involved, so the same checks give the same answer every run.
  • pass: Frontmatter block present
  • pass: Frontmatter declares a name
  • pass: Frontmatter declares a description
  • pass: Size between 200 bytes and 200 KB (2177 bytes)
  • pass: No zero-width or bidi control characters
  • pass: No instruction hidden inside an HTML comment
  • pass: No link to an exfiltration or paste host
  • pass: No credential-shaped string
  • pass: No instruction to send local credentials anywhere
  • pass: No text hidden with inline styles
  • pass: No prompt-injection phrasing
  • pass: No curl or wget piped into a shell
  • pass: No recursive delete of root, home or parent
  • pass: No instruction to read or print local credentials
  • pass: No base64 blob over 200 characters
  • pass: No link to a raw IP address
  • pass: No script tag

Source

GitHub

tongminglaic/ako4x · 84 stars · license MIT · pushed 2026-09-15 · branch main

API

GET https://markdownregistry.com/api/v1/artifacts/art_dghrhj3h4zfsfwyg
GET https://markdownregistry.com/api/v1/resolve?ref=tongminglaic/ako4x/profiler-ncu
GET https://markdownregistry.com/api/v1/blob/671d0c65b298e0e6f0f8161ccc098f7651ba3d03db6fb4589f79c2754139bbe2

Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.

Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.

More from tongminglaic/ako4x

CLAUDE.md CLAUDE.md file
tongminglaic/ako4x · CLAUDE.md
git:20260531.0fd4b5f · audit A · 84 stars
CLAUDE.md@scripts CLAUDE.md file
tongminglaic/ako4x · scripts/CLAUDE.md
git:20260531.0fd4b5f · audit A · 84 stars
bench SKILL.md skill
tongminglaic/ako4x · templates/skills/bench/SKILL.md · Run performance benchmarks to get a verdict on whether a kernel change actually helped. Use --ab-compare for sub-1x…
git:20260531.0fd4b5f · audit A · 84 stars
benchmark SKILL.md skill
tongminglaic/ako4x · templates/skills/benchmark/SKILL.md · Reference for the active benchmark harness — what it IS and how it behaves (the active benchmark is flashinfer-bench)…
git:20260531.0fd4b5f · audit A · 84 stars
cpp SKILL.md skill
tongminglaic/ako4x · templates/skills/cpp/SKILL.md · Host-side C++ kernels (`[build].language = cpp`) — the `solution/{kernel.cpp, binding.py}` two-file TVM-FFI binding…
git:20260531.0fd4b5f · audit A · 84 stars
cuda SKILL.md skill
tongminglaic/ako4x · templates/skills/cuda/SKILL.md · CUDA C++ (.cu) kernel reference — TVM-FFI direct-export vs Python-binding entry points, the chevron-launch null-stream…
git:20260531.0fd4b5f · audit A · 84 stars
cute-dsl SKILL.md skill
tongminglaic/ako4x · templates/skills/cute-dsl/SKILL.md · CuTe DSL reference — NVIDIA's CUTLASS Python DSL. Use whenever writing or debugging a CuTe DSL kernel — covers the…
git:20260531.0fd4b5f · audit A · 84 stars
sanitizer SKILL.md skill
tongminglaic/ako4x · templates/skills/sanitizer/SKILL.md · Run NVIDIA compute-sanitizer (memcheck / racecheck / initcheck / synccheck). FIRST action on `INCORRECT_NUMERICAL` or…
git:20260531.0fd4b5f · audit A · 84 stars
tilelang SKILL.md skill
tongminglaic/ako4x · templates/skills/tilelang/SKILL.md · TileLang DSL reference — the `@tilelang.jit` factory + `@T.prim_func` pattern, and the TileLang PDL binding…
git:20260531.0fd4b5f · audit A · 84 stars
triton SKILL.md skill
tongminglaic/ako4x · templates/skills/triton/SKILL.md · Triton DSL reference for kernel writers — num_warps/num_stages choice, the small-N fp8 MMA-throughput regression…
git:20260531.0fd4b5f · audit A · 84 stars

Every file in tongminglaic/ako4x

Browse by kind, by grade A, or by owner.