OCuLink eGPU Home Lab

This article summarizes a hardware project: adding a desktop-class GPU to a Lenovo ThinkCentre mini PC using an M.2 → OCuLink adapter and an OCuLink eGPU enclosure. For step-by-step detail, see the linked notes below.

Mini PCs are efficient and quiet, but integrated graphics hit a wall for gaming, CUDA, local inference, and transcoding. Thunderbolt eGPUs are convenient, yet they share a single high-speed link and often show a larger performance gap than a native PCIe run. OCuLink exposes PCIe lanes to an external box, so in practice you are much closer to “GPU on PCIe ×4” than to a typical TB enclosure—at the cost of opening the host, routing a cable out of the case, and giving up an M.2 slot (often Wi‑Fi or a secondary NVMe).

Hardware used in this build

Role Part
Host Lenovo ThinkCentre mini PC (AMD Ryzen 5 PRO, ~30 GB RAM)
Link M.2 (M-key) → OCuLink adapter
Enclosure AOOSTAR (or similar) OCuLink eGPU box
GPU ASUS GeForce RTX (desktop card)
PSU Lite-On FS9323 (in enclosure)
Cable OCuLink ↔ OCuLink (prefer a short, high-quality cable; avoid sharp bends)

Build flow (high level)

  1. Prepare — Back up, unplug AC, ground yourself, photo screw layouts. Plan a backup network path (Ethernet or USB Wi‑Fi) if the adapter uses the Wi‑Fi M.2 slot.
  2. Mini PC — Open the case, seat the M.2 OCuLink board, route the pigtail clear of the CPU fan and heatsink, exit the cable through a vent or planned gap, then close without pinching wires.
  3. Enclosure — Mount the PSU, install the GPU until the slot clicks, connect PCIe power (6+2 / 8-pin as required), keep airflow unobstructed.
  4. Power sequence — Enclosure on first, then the mini PC (helps reliable enumeration).
  5. BIOS — Look for Above 4G decoding, match PCIe generation to what your adapter supports, and be ready to toggle ReBAR while debugging.
  6. OS — On Linux: lspci, vendor tools such as nvidia-smi, and a monitor like btop to confirm VRAM and thermals. On Windows: Device Manager + vendor driver package.

ASUS GeForce RTX mounted in OCuLink eGPU enclosure

Internal view: OCuLink adapter and cable routing inside the mini PC

Complete desk setup with ThinkCentre mini PC and external GPU enclosure

btop system monitor showing CPU, memory, and GPU stats

Expectations and caveats

  • Performance: Plan for roughly ~5–10% loss versus the same GPU in a desktop ×16 slot—often still better than many Thunderbolt setups for bandwidth-sensitive work.
  • Footprint: You now have two powered units and a fat data cable on the desk.
  • Risk: Internal mods can void warranty; bad PSU choice or pinched cables can damage hardware.

Do this at your own risk. Use a PSU rated for your GPU, keep vents clear, and monitor temperatures on both the mini PC and the enclosure under load.

Who this is for

Good fit: You already want the mini PC as the always-on brain, and you want one strong GPU for ML, builds, games, or transcode, and you are comfortable with tinkerer-grade integration.

Poor fit: You need a single travel bag solution, you cannot lose internal Wi‑Fi, or you must keep vendor support / warranty untouched.

Further reading

Quick Linux checks 

lspci | grep -i -E 'nvidia|vga|3d'
nvidia-smi

Last updated: April 2026