Skip to main content
affpapa.org
RU EN
skill CX Codex

tilegym-improve-cutile-kernel-perf — Claude Code skill: setup guide

Systematically profiles, tunes, and benchmarks cuTile GPU kernels in TileGym. Runs iterations on real hardware to lower latency while preserving accuracy.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓ 
npx skills add NVIDIA/skills --skill tilegym-improve-cutile-kernel-perf -a claude-code
Open on GitHub

What tilegym-improve-cutile-kernel-perf does

The skill establishes a structured tuning workflow for cuTile kernels inside the TileGym repository. It starts by checking arithmetic intensity to classify the kernel as memory-bound or compute-heavy. The assistant then isolates changes to one parameter per run, testing tile dimensions, TMA usage, occupancy, latency hints, or persistent scheduling.

Every run executes validation checks and benchmarks through pytest, logging latency figures directly into @sandbox/perf_results.md. If output accuracy fails or timing degrades by more than 2% relative to baseline, the change is reverted immediately.

Benchmarking requires an active GPU node equipped with Blackwell, Hopper, or Ampere hardware. Running this workflow on a machine without matching GPUs is pointless because the profiling loop depends on actual execution timings.

What it helps with

  • reduce cuTile kernel execution latency in milliseconds
  • tune tile sizes and persistent scheduling configurations
  • verify numerical correctness through pytest after each code change
  • log benchmark iterations and speedups in @sandbox/perf_results.md

Example prompts for the agent

  • Optimize the cuTile kernel in src/tilegym/ops/cutile/ and benchmark latency
  • Tune tile sizes and test persistent scheduling for this cuTile operator
  • Profile this kernel and log performance changes to sandbox

How to install tilegym-improve-cutile-kernel-perf

1Skills CLI (Vercel)

npx skills add NVIDIA/skills --skill tilegym-improve-cutile-kernel-perf -a claude-code

2Manually: put the files in place

~/.claude/skills/tilegym-improve-cutile-kernel-perf/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/tilegym-improve-cutile-kernel-perf/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about tilegym-improve-cutile-kernel-perf

How to install tilegym-improve-cutile-kernel-perf in Claude Code?

Run npx skills add NVIDIA/skills --skill tilegym-improve-cutile-kernel-perf -a claude-code in your terminal.

What does tilegym-improve-cutile-kernel-perf do?

It guides the agent through an iterative loop of profiling, applying single-parameter cuTile optimizations, verifying numerical correctness, and logging latency.

What hardware is required for this skill?

The benchmarking step requires an NVIDIA GPU node with Blackwell, Hopper, or Ampere architecture to collect real runtime metrics.

Similar skills

Everything for AI and ML →
cuda-debugging mohitmishra786 Walks the agent through CUDA debugging: cuda-gdb, Compute Sanitizer, GPU core dumps, error codes 700 and 702. Useful when a kernel crashes or returns wrong values. CX Codex +1 301installs web-perf cloudflare Audit Core Web Vitals, diagnose rendering bottlenecks, and inspect network dependencies directly through your agent. CX Codex +1 110,200installs perf-investigation-logger composio-community Appends structured performance investigation notes, quotes, and evidence to project logs. Keeps benchmarks and optimization decisions organized. CX Codex +1 1,600installs perf-profiler composio-community Runs profilers against CPU and memory hot paths, captures flame graphs, and records file and line evidence. CX Codex +1 1,600installs perf-lighthouse tech-leads-club Runs local Lighthouse audits via CLI or Node API, parses reports, and configures performance budgets. Useful for tracking Core Web Vitals and integrating speed checks into CI. CX Codex +1 320installs profile-model modular Profiles MAX models to check GPU utilization and identify the GPU kernels taking the most time. Use it when inference is slow or a trace needs inspection. CX Codex +1 1,200installs

From the same set178

hsb-test Runs automated QA test plans against Holoscan Sensor Bridge hardware on a remote devkit. It filters tests from your document, executes them in a demo container, and writes a results report. CX Codex +1 1,700installs hsb-setup Sets up a Holoscan Sensor Bridge environment by connecting to a devkit, building a container, and checking board connectivity. Use it for first HSB demo bring-up. CX Codex +1 1,700installs tao-run-on-slurm Submit TAO jobs to a remote SLURM GPU cluster over SSH, with Pyxis/Enroot containers. Use it when training or inference needs cluster storage on Lustre. CX Codex +1 1,600installs tao-run-on-kubernetes Turns a TAO spec-bundle into a Kubernetes Job and submits it with plain kubectl. For training runs and inference on EKS, GKE, AKS or an on-prem GPU cluster. CX Codex +1 1,600installs tao-run-inference-service Deploys network-specific NVIDIA TAO inference microservices in containers and halts them. Use it to run checkpoints and query endpoints without cloud storage. CX Codex +1 1,600installs tao-validate-dataset-format This skill runs tao-daft validate on NVIDIA TAO DAFT data and helps interpret the results. Use it to check dataset structure, schemas, and cross-references. CX Codex +1 1,600installs tao-train-mask-auto-encoder A skill that runs Masked Auto-Encoder training in TAO: random patches get masked and rebuilt to learn visual representations. CX Codex +1 1,600installs tao-train-single-step Runs standard train, eval, and export workflows for NVIDIA TAO models without AutoML loops. Built for single-run fine-tuning on labeled datasets. CX Codex +1 1,600installs

Project sponsors

Companies that keep this analytics open
Gold sponsor
Nutra.Club

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor