Skip to content

Validated System Snapshot

This document records the sanitized snapshot of the live validated Ubuntu + k3s + NVIDIA GPU environment reconciled on 2026-07-20.

1. System & Host Environment

Parameter Observed Value / Setting
Host System waqasm86-ThinkPad-T450s (Ubuntu Linux 6.8+ kernel)
Kubernetes Substrate k3s single-node control plane & worker
Container Runtime containerd with nvidia-container-runtime OCI hook
NVIDIA GPU Hardware NVIDIA GPU (GeForce 940M / CUDA 12.8+ Driver 535+)
Allocatable GPU Resource 1x nvidia.com/gpu
RuntimeClass RuntimeClass/nvidia registered and active

2. Live Kubernetes Workload Inventory (kubectl get all -A)

gpu-operator Infrastructure Substrate

NAMESPACE     NAME                                                            READY   STATUS      RESTARTS   AGE
gpu-operator  pod/gpu-operator-7c46869546-v5v9j                              1/1     Running     8          11h
gpu-operator  pod/k3s-nvidia-edge-node-feature-discovery-gc-6576ddb78b-gnv25 1/1     Running     2          11h
gpu-operator  pod/k3s-nvidia-edge-node-feature-discovery-master-6bff59d775   1/1     Running     5          11h
gpu-operator  pod/k3s-nvidia-edge-node-feature-discovery-worker-tmt78        1/1     Running     3          11h
gpu-operator  pod/nvidia-container-toolkit-daemonset-nnp9q                   1/1     Running     3          10h
gpu-operator  pod/nvidia-cuda-validator-ksh5k                                0/1     Completed   0          2m13s
gpu-operator  pod/nvidia-dcgm-exporter-nh6xm                                 1/1     Running     3          10h
gpu-operator  pod/nvidia-device-plugin-daemonset-mghzn                       1/1     Running     3          10h
gpu-operator  pod/nvidia-operator-validator-qrt45                            1/1     Running     2          10h

llm-observability Workloads & Telemetry

NAMESPACE          NAME                                                           READY   STATUS    RESTARTS   AGE
llm-observability  pod/alertmanager-kube-prometheus-stack-alertmanager-0           2/2     Running   4          8h
llm-observability  pod/kube-prometheus-stack-operator-677cd9856c-94hwd             1/1     Running   2          8h
llm-observability  pod/llm-observability-stack-grafana-5687d5cb96-kwvrk            3/3     Running   12         8h
llm-observability  pod/llm-observability-stack-kube-state-metrics-6557cd9478-4t5h7 1/1     Running   2          8h
llm-observability  pod/llm-observability-stack-prometheus-node-exporter-th7vj      1/1     Running   2          8h
llm-observability  pod/ollama-5f96f99b68-88h6s                                     1/1     Running   0          3m
llm-observability  pod/open-webui-0                                                1/1     Running   2          8h
llm-observability  pod/open-webui-redis-5b4b5b48d5-xkg4f                           1/1     Running   2          8h
llm-observability  pod/opentelemetry-collector-7dd4d89978-d76xl                    1/1     Running   2          8h
llm-observability  pod/opentelemetry-operator-74f6c5bfbd-tcpnd                     1/1     Running   2          8h
llm-observability  pod/prometheus-kube-prometheus-stack-prometheus-0               2/2     Running   4          8h

3. Active Kubernetes Services & Exposed Ports

Namespace Service Name Type Cluster-IP Ports Role
gpu-operator gpu-operator ClusterIP 10.43.167.181 8080/TCP Operator metrics
gpu-operator nvidia-dcgm-exporter ClusterIP 10.43.201.102 9400/TCP GPU telemetry
llm-observability ollama ClusterIP 10.43.231.195 11434/TCP Ollama GGUF API
llm-observability open-webui ClusterIP 10.43.6.112 8080/TCP Open WebUI Frontend
llm-observability open-webui-redis ClusterIP 10.43.162.152 6379/TCP Session/Cache Store
llm-observability opentelemetry-collector ClusterIP 10.43.202.227 8888, 4317, 4318 OTLP Collector
llm-observability llm-observability-stack-grafana ClusterIP 10.43.169.54 80/TCP Grafana Dashboards
llm-observability kube-prometheus-stack-prometheus ClusterIP 10.43.250.236 9090/TCP Prometheus Engine
llm-observability kube-prometheus-stack-alertmanager ClusterIP 10.43.152.143 9093/TCP Alerting System

4. Verification & Validation Summary

  • Substrate Health: nvidia-cuda-validator job executed to completion (Completed), verifying CUDA execution inside container pods.
  • Node Feature Discovery: nfd-master and nfd-worker actively label node features.
  • Model Residency: Ollama running with local GGUF model (gemma3-1b-it-gguf-local / llama3.2:1b), offloading layers under the 900 MiB VRAM operational ceiling.
  • Telemetry Pipeline: DCGM Exporter metrics successfully ingested into Prometheus and visualizable in Grafana edge dashboards.