# Gpu Workload Troubleshooter

> - Decision-tree diagnosis for GPU workloads on Kubernetes: pods stuck Pending with Insufficient nvidia.com/gpu, FailedScheduling, missing device plugin signatures, nodes that launch but never register GPUs, node pools that won't scale, unhealthy GPU hardware (AcceleratedHardwareReady false), CrashLoopBackOff during slow model loads, and OOMKilled vs CUDA out-of-memory. Use whenever the user reports a pod pending, GPU not detected, "0/N nodes are available", nvidia.com/gpu not allocatable, scheduling failures, taint/toleration mismatches, node autoscaler not reacting, NodeClaim stuck, gpu resource not registered, exit code 137, liveness probe killing an inference server, CUDA OOM, or "it worked yesterday". Includes kubectl commands per step. For designing (rather than debugging) autoscaling use sibling gpu-autoscaling-engineer; for sharing semantics use gpu-sharing-advisor; for quota rejections at creation see researcher-tenancy-provisioner.

## Facts
- Page: https://tashan.sh/capability/skill-cloud-byte-consulting-gpu-workload-troubleshooter
- tashan id: skill:Cloud-Byte-Consulting/gpu-workload-troubleshooter
- Source: https://github.com/Cloud-Byte-Consulting/plugins
- Type: skill
- Category: other
- tashan score: not scored (catalogued only — too little public evidence)
- Adoption: 9.0
- Upkeep: 95.0
- Freshness: 90.0
- Evidence coverage: 84% of the inputs this score can use
- Health: active
- Instruction depth: not yet graded
- License: Apache-2.0
- Official: no

## Install

```sh
cp -r gpu-workload-troubleshooter ~/.claude/skills/
```

## Security audit
Not scanned. We audit npm-published capabilities; this one has no npm package we can resolve, or has not reached the queue. This is not a clean bill of health.

---
Measured 2026-08-22 by tashan (https://tashan.sh) from public evidence. Scorer s5.