agentsclimarketplace

Gpu host tuning

Skill air-gapped/skills/.claude/skills/gpu-host-tuning

Claude Code plugin marketplace — 58 installable reference skills across vLLM/SGLang inference, Kubernetes & Harvester, GPU host bring-up, observability, security, and agent workflows.

Install
npx -y skills add air-gapped/skills --skill gpu-host-tuning

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Audit AND tune Linux/GPU inference hosts — read-only host snapshot (CPU power state, C-states, NUMA topology, PCIe link state, GPU settings, kernel boot params, sysctl, ulimits, IRQ affinity, container runtime), optional pinned-host↔GPU memcpy bench (torch + numactl), and per-lever cheat-sheets to flip settings (governor, EPP, cpuidle, persistence, ECC, hugepages, intel_iommu, NCCL env, tuned-adm profiles, Dell/Supermicro/HPE BIOS guidance). Sits beneath any inference framework (vLLM, sglang, TensorRT-LLM) — about the host, not the framework.

SKILL.md

5.9 KB, as published. Nobody here has run it

gpu-host-tuning

Host-side tuning + audit for Linux GPU inference servers. Sits beneath any inference framework (vLLM, sglang, TensorRT-LLM, llama.cpp).

Assumes a working host. Driver, nvidia-fabricmanager, and NVLink/NVSwitch must already be healthy — if nvidia-smi doesn't list every GPU, or fabricmanager isn't active (running), that is a bring-up problem, not a tuning one: use nvidia-datacenter-bringup (same inference-host plugin) first. Tuning a half-initialised host produces numbers that describe the fault, not the hardware.

Three modes:

  1. Audit — read-only snapshot
  2. Bench — ground-truth pinned-host↔GPU memcpy ceiling
  3. Tune — apply individual levers from the cheat-sheet

This file is a pointer map. The actual logic lives in scripts/ and the authoritative references in references/.

Quick start

# From the skill directory — typically ~/.claude/skills/gpu-host-tuning
# (personal) or .claude/skills/gpu-host-tuning (project install).

# Audit (read-only, ~60s)
./scripts/collect.sh

# Audit + pinned-memcpy bench (needs torch + CUDA, ~5 min)
./scripts/collect.sh --bench

The script prompts for the output parent dir on first interactive run and remembers the choice. Override via --out <dir> or HOST_AUDIT_DIR=<dir>. Default snapshot dirname is gpu-host-tuning-<host>-<UTC>.

What the snapshot captures

One file per probe, numbered by section. See references/probe-interpretation.md for the full file-by-file decoder.

SectionWhat
00-09 metacollector version, run timestamp, args
10-19 system + firmwaredmidecode (BIOS, CPU, memory DIMMs), lshw, /sys/class/dmi
20-29 CPU + power + C-statesgovernor, EPP, intel_pstate / amd_pstate, cpuidle states + disable mask, turbostat 5s residency, microcode, vulnerabilities, thermal zones
30-39 memory + NUMAnumactl -H, /proc/meminfo, THP, numa_balancing, vm tunables, hugepages
40-49 kernel + limitsuname, /etc/os-release, /proc/cmdline, sysctl -a, ulimit, /sys/devices/system/cpu/vulnerabilities, dmesg, IRQ affinity, env vars in vllm processes
50-59 PCIelspci tree + verbose, AER counters, link width/speed for every NVIDIA device
60-69 GPUnvidia-smi -q full, topo -m, nvlink --status, clocks/power/ECC, dmon 5s, dcgmi diag
70-79 networkNICs, IB (ibstat / ibv_devinfo), ethtool ring sizes, RDMA links
80-89 storagelsblk, NVMe id-ctrl, smartctl, mount flags, io scheduler
90-99 container runtimecontainerd version, CDI specs, cgroup v2, kubelet config, RKE2 config

Three modes — what each maps to

ModeWhatReference
Audit./scripts/collect.sh writes the snapshot directoryreferences/probe-interpretation.md decodes each numbered file
Bench./scripts/collect.sh --bench adds the pinned-memcpy CSVreferences/session-findings.md lists baselines per chassis
TuneNo script — apply individual levers from the cheat-sheetreferences/recommended-tunings.md (lever-by-lever) and references/tuned-profiles.md (apply via tuned-adm)

When to use which reference

GoalRead
Apply NVIDIA's stock DGX tuningsreferences/tuned-profiles.md
See exactly what NVIDIA's settings packages flip (per-platform JSON, GRUB drop-ins, sysctl, units)references/nvidia-dgx-config-decoder.md
Run a proper bring-up flowreferences/bringup-recipe.md
Find the lever the audit flaggedreferences/recommended-tunings.md
Decode an audit output filereferences/probe-interpretation.md
Tune a Dell XE9680 (H100/H200, SPR/EMR)references/dell-xe9680.md
Tune a Dell XE9780 / XE9780L (B200/B300, Granite Rapids)references/dell-xe9780.md
Understand why cpufreq/cpuidle is empty inside a cloud VMreferences/virt-and-cloud-quirks.md
See measured baselines from real boxesreferences/session-findings.md

Comparing two snapshots

Two snapshots on the same host (e.g., pre-tune and post-tune) can be compared with diff -ruN snap_pre/ snap_post/. For a structured impact ranking, use references/probe-interpretation.md to interpret deltas.

Companion skills

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.