Every premium and members-only resource for $9.99/month, as a no-commitment subscription. See the plans

PalateKit

Menu

Request a resource

Describe the resource you are missing in one sentence. We will email you when it is published.

0 / 280

Unlock PalateKit

Unlock everything

Plan

Paying means accepting our terms of sale.

Secure payment by Stripe. Receipts and invoices sent by email.

Welcome back

Sign in to install, download or save this resource. A free account is enough for free resources.

Iterative cuTile kernel optimization v1.0.0

Use this resource to iteratively improve the performance of a cuTile kernel in TileGym

New

  • 389 downloads
  • 117 likes
  • 141 saves

Free

Install with the CLI

In a terminal, from the folder you want:

npx --yes palatekit@latest install cutile-kernel-performance-optimization

With no option, the skill goes to ~/.agents/skills/; add --agent claude-code (or codex, cursor…) to target an agent.

First time: connect the CLI with

npx --yes palatekit@latest login

The agent creates an experiment branch, sets a baseline, then loops: change one knob, measure, log. It covers tile sizes, occupancy, autotune, TMA, persistent scheduling, num_ctas and IR analysis. You get an experiment log and a faster kernel.

Mostly picked by

Updated November 2, 2026

Card for the “Iterative cuTile kernel optimization” skill

Details

Generates with
Your agent's model
Licence
CC-BY-4.0

Similar resources

View all

What you get

  • references/optimization-playbook.md: optimization method
  • references/perf-knobs-catalog.md: knob catalog
  • references/ir-dump-guide.md: IR reading guide

Good fit for

  • A TileGym kernel is slower than expected
  • You want every tuning trial recorded in a log

Not for

    Tags

    • cuTile
    • optimization
    • GPU
    • profiling

    About

    You say

    “Optimize the cuTile kernel my_op: measure the baseline, then try different tile sizes.”

    Card for the “Iterative cuTile kernel optimization” skill

    Skill published by nvidia under the CC-BY-4.0 license, delivered in its original version.

    Show more

    How to use it

    1. Install the skill in your agent's skills folder
    2. Ask your agent to iteratively improve the performance of a cuTile kernel in TileGym
    3. Review and adjust the result

    Content preview

    Files

    • BENCHMARK.md3.6 KB
    • evals/evals.json5.3 KB
    • LICENSE11.3 KB
    • NOTICE-PALATEKIT.md0.5 KB
    • references/cutile-api-reference.md31.3 KB
    • references/cutile-patterns-reference.md7.8 KB
    • references/ir-dump-guide.md8.1 KB
    • references/optimization-playbook.md15.2 KB
    • references/perf-knobs-catalog.md6.7 KB
    • references/performance-model.md19.2 KB
    • skill-card.md4.3 KB
    • SKILL.md7.8 KB
    • skill.oms.sig6 KB

    Origin and licence

    Listed unmodified

    Author
    nvidia
    Source
    github.com/nvidia/skills/tree/ca116d60beafe1b4459f0ab790b7b450912c56b0/skills/tilegym-improve-cutile-kernel-perf
    Licence
    CC-BY-4.0

    Notice

    « tilegym-improve-cutile-kernel-perf » (nvidia, CC-BY-4.0) livré dans sa version d'origine : fiche en français et en anglais par PalateKit.

    User reviews

    –

    No ratings yet

    Loading reviews…

    Delete your review?

    Your rating and comment will be removed from the page.