Dev & Engineering ✓ NVIDIA · Official cutilegpu-kernelperformance-optimizationprofilingautotuningcudatilegym

TileGym cuTile Kernel Optimizer

Systematically reduce cuTile GPU kernel latency through profiling, bottleneck diagnosis, and controlled tuning.

FollowSkills review · FSRS-2.0
Not recommended
44/ 100 5-point scale 2.2 / 5
Trust12 / 25 · 2.4/5

The skill constrains editable files, requires correctness checks, and explicitly warns that out-of-bounds TMA use on Ampere can silently corrupt results. However, it includes git checkout/reset, commits, and rm -rf operations, while user confirmation, protection of pre-existing uncommitted changes, rollback boundaries, and data-flow disclosure are incomplete. The repository license metadata is NOASSERTION and conflicts with the file's declared licenses, so points were deducted.

Reliability6 / 20 · 1.5/5

The three-phase workflow, one-change-per-iteration method, experiment log, benchmark commands, and failure diagnostics are reasonably coherent. Reliability is reduced by dependence on specific GPUs, cuda-tile, tileiras, repository tests, and external environment setup; placeholder paths and commands are not directly reproducible, and the Blackwell/Hopper/Ampere support statements are inconsistent. Static calibration limits this to 6.

Adaptability8 / 15 · 2.7/5

The audience, cuTile performance scenario, trigger phrases, kernel locations, and optimization scope are fairly clear, with some non-fit cases documented. However, systematic non-goals, output expectations, and fallback behavior when prerequisites fail are incomplete; there is no Chinese-language guidance, and the GPU-node requirement limits general applicability. Points were deducted.

Convention8 / 15 · 2.7/5

The documentation is substantially layered across API, patterns, performance model, playbook, logging, version, and author metadata. It still contains duplicated material, references files not included in the supplied evidence, lacks a clear changelog and maintenance/update path, uses a future-dated version format without release context, and conflicts with the repository license metadata. Points were deducted.

Effectiveness6 / 15 · 2.0/5

The skill gives a directly usable workflow for profiling, applying one optimization at a time, checking correctness, comparing latency, and deciding keep/revert, with an explicit tracking-table format. Its value depends heavily on an available GPU, complete TileGym tests, and real benchmarks; claimed impact examples are not independently verified in the supplied files, and the included evaluation report has an overall FAIL verdict. Static calibration therefore caps and reduces the score to 6.

Verifiability4 / 10 · 2.0/5

The committed evaluation tasks, reported results, command templates, performance-log schema, and diagnostic procedures provide some auditability. There is no real CI workflow, independently checkable execution output, or corroborating reproduction evidence; the benchmark is primarily a report claim, and key paths were not executed during this review. Static calibration supports 4 points.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 55f18499943e
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Before execution, explicitly confirm branch creation, commits, resets, and temporary-directory cleanup, and protect existing uncommitted changes.
  • Verify the actual GPU architecture, cuTile/tileiras versions, test paths, and TMA out-of-bounds semantics; in particular, prevent silent corruption on Ampere by using ZERO padding or a safe fallback.
  • Resolve the license metadata conflict, add real CI/test reproduction evidence, and rerun evaluation; the supplied benchmark reports an overall FAIL.
  • The documentation has no Chinese guidance and the core workflow depends on specific GPUs and NVIDIA software infrastructure.
See the full review method →

What does this skill do, and when should you use it?

This skill is for optimizing cuTile GPU kernels in the TileGym project. It establishes a baseline, applies one optimization per experiment, verifies correctness, and reruns performance benchmarks. It covers tile sizes, occupancy, autotuning configurations, TMA, latency hints, persistent scheduling, num_ctas, and flush_to_zero. Latency is the core metric, while numerical correctness must be preserved throughout.

It guides an agent to create an optimization branch, locate the TileGym cuTile kernel, inspect its launch wrapper, registration, and autotune configurations, and consult the referenced optimization materials. It creates and updates @sandbox/perf_results.md, runs pytest correctness and performance commands, compares latency_ms across iterations, and decides whether to keep, revert, or stop. It can also use IR-level inspection and bottleneck analysis to diagnose performance or compilation issues.

  1. A TileGym developer needs to reduce the GPU latency of an existing cuTile kernel.
  2. An engineer needs profiling-guided diagnosis of memory, compute, occupancy, or scheduling bottlenecks.
  3. A kernel author wants to evaluate TMA, tile-size, occupancy, or persistent-scheduling changes one at a time.
  4. A team needs reproducible records of correctness results, latency measurements, and optimization decisions.
  5. A developer needs to compare generated IR while investigating a cuTile performance or compilation issue.

What are this skill's strengths and limitations?

Pros
  • Provides a concrete baseline-to-experiment-to-decision workflow.
  • Uses latency (ms) as the primary metric and requires correctness verification for every change.
  • Covers specific cuTile tuning areas, including TMA, tile sizes, occupancy, autotuning, and IR debugging.
  • Requires structured records of latency, correctness, and iteration status.
Limitations
  • Requires the TileGym repository, cuTile, Python pytest, Git, and a GPU node.
  • Benchmarking is documented for Blackwell, Hopper, or Ampere GPUs, with no alternative CPU-only workflow.
  • The skill provides no kernel-specific test data or demonstrated performance-improvement results.
  • A ten-minute limit per iteration and a 25-iteration cap may constrain complex optimization work.

How do you install this skill?

Install the individual skill from NVIDIA/skills with the skills CLI:

npx skills add nvidia/skills --skill tilegym-improve-cutile-kernel-perf --yes

You may also use --agent codex or --agent claude-code to target a client. The README does not document a manual folder-copy installation path.

How do you use this skill?

After the skill is loaded, use a request such as:

"Optimize this cuTile kernel performance in TileGym."

Other supported trigger wording includes "improve kernel perf", "tune cutile performance", and "make kernel faster". Prepare the TileGym repository, a suitable GPU node, and the test environment; the workflow asks for confirmation during setup before experimentation begins.

FAQ

Which GPUs does the workflow require?
The source requires a GPU node for benchmarking and names Blackwell, Hopper, and Ampere GPUs.
Can the optimization change numerical behavior?
No. The workflow prohibits changing functional semantics and requires correctness testing after every optimization; failures should be reverted immediately.
When does the experiment loop stop?
It stops when performance goals are met, gains plateau, attempts are exhausted, more than 25 iterations occur, or the user interrupts it. Two consecutive iterations without improvement are also a stop condition.

More skills from this repository

All from NVIDIA/skills

Dev & Engineering ✓ NVIDIA · Official

CuTile Autotuning Guide

Design, implement, and validate low-overhead autotuning for CuTile GPU kernels.

Dev & Engineering ✓ NVIDIA · Official

TileGym cuTile Kernel Guide

A step-by-step guide for registering, implementing, testing, and benchmarking a new cuTile GPU kernel in TileGym.

Dev & Engineering ✓ NVIDIA · Official

cuTile Python Kernel Assistant

Guides developers in writing, debugging, validating, and optimizing high-performance cuTile GPU kernels.

Dev & Engineering ✓ NVIDIA · Official

cuTile-to-Triton Kernel Converter

Port cuTile GPU kernels to validated, tunable Triton implementations.

Dev & Engineering ✓ NVIDIA · Official

cuTile Python-to-Julia Converter

Port cuTile Python GPU kernels into tested cuTile.jl Julia implementations.

Automation & Ops ✓ NVIDIA · Official

cuOpt Installation Guide

Install and verify NVIDIA cuOpt for Python, C, or REST deployment.

Dev & Engineering ✓ NVIDIA · Official

NeMo MBridge GPU Memory Tuning

Diagnose GPU OOMs in Megatron Bridge training and reduce fragmentation, activation, and PEFT memory usage.

Dev & Engineering ✓ NVIDIA · Official

cuPyNumeric Migration Readiness

Assess whether NumPy code is ready to scale on GPUs before committing to a substantial cuPyNumeric port.

Dev & Engineering ✓ NVIDIA · Official

DOCA GPUNetIO Development Skill

Helps developers connect CUDA kernels on NVIDIA GPUs to DOCA network queues for GPU-side packet I/O and debugging.

Automation & Ops ✓ NVIDIA · Official

TAO NVIDIA GPU Host Setup

Checks and standardizes NVIDIA drivers, CUDA, and container runtime prerequisites for TAO GPU hosts.

Dev & Engineering ✓ NVIDIA · Official

NVIDIA Skill Finder

Finds the right NVIDIA agent skill for product, hardware, and workflow requests.

Dev & Engineering ✓ NVIDIA · Official

DOCA GPI GPU-Initiated RDMA Skill

Helps CUDA kernels drive RDMA queues directly from GPU memory without host-CPU mediation.

Dev & Engineering ✓ NVIDIA · Official

GPUNetIO RDMA Write Latency Benchmark

Guides reproducible measurement of CUDA-kernel-initiated GPUNetIO RDMA WRITE latency and tail behavior.

Automation & Ops ✓ NVIDIA · Official

Holoscan NGC Container Installer

Install, validate, and launch NVIDIA Holoscan SDK through an NGC Docker container.

Dev & Engineering ✓ NVIDIA · Official

Holoscan Native Ubuntu Installer

Install and verify the Holoscan SDK C++ environment on Ubuntu from NVIDIA’s apt repository.

Dev & Engineering ✓ NVIDIA · Official

Catheter Navigation Setup Verification

Checks host, GPU, and Python path readiness for catheter navigation.

Dev & Engineering ✓ NVIDIA · Official

TileGym Transformers Kernel Integration

Integrate TileGym kernels into Hugging Face Transformers models without editing Transformers source.

Dev & Engineering ✓ NVIDIA · Official

Holoscan Python Wheel Installer

Install Holoscan SDK Python bindings in a virtual environment and verify them on a Linux NVIDIA GPU system.

Dev & Engineering ✓ NVIDIA · Official

Holoscan SDK Source Builder

Build Holoscan SDK from source in a Docker container and produce a local CMake-consumable installation tree.

Data & Analysis ✓ NVIDIA · Official

Synthetic Brain MRI Generator

Generate synthetic brain MRI volumes through NVIDIA’s documented workflow.

Related skills