Norwegian reference for local AI — registers, plans and honest tests.

draft · skeleton

01 · Jetson Orin Nano Super

Goal: a small NVIDIA Jetson that runs local AI for hobby use. This page is a skeleton — not a finished, measured guide.

Last updated:

No measurements yet. Tokens/s, noise and power are “not measured yet”. Do not treat forum numbers as ours.

What you build

A Jetson Orin Nano Super node with storage enough for a few GGUF models, cooling that works in a living room, and a runtime (typically Ollama or llama.cpp) that actually uses the GPU.

Parts list (BOM)

Prices are “unknown” unless taken from an official page in this round. Links are starting points — check stock and region yourself.

PartWhyLinkPrice
Jetson Orin Nano Super Developer Kit (or Super module + carrier) Compute with GPU / unified memory for local inference NVIDIA Jetson Orin unknown
microSD or NVMe (recommended for models) OS + model files; NVMe is faster for large GGUF unknown — choose by kit variant unknown
Power supply as required by the kit Stable operation under load Follow kit docs unknown
Optional: active cooling / case Noise and temperature in a living room unknown unknown

Steps draft

  1. Note kit variant, JetPack/L4T version and storage choice before you flash.
  2. Install OS/JetPack following NVIDIA’s instructions for your variant.
  3. Set up storage (preferably NVMe) and move the model directory there.
  4. Install a runtime (Ollama or build llama.cpp with the right CUDA/sm_*).
  5. Start with a small model (3B–4B class in Q4). Do not jump to large weights.
  6. Write down tokens/s, memory use, noise and whether the GPU path is actually active.

Measurements

not measured yet

Comments

Comments load only when you click. They are stored in GitHub Discussions (USA) via Giscus. See privacy.

Giscus is not connected yet. Repo, category ID and mapping are placeholders — no third-party script loads.