cd /news/ai-infrastructure/homelab-critique · home topics ai-infrastructure article
[ARTICLE · art-130632] src=forum.level1techs.com ↗ pub= topic=ai-infrastructure verified=true sentiment=· neutral

Homelab critique

A homelab user identified as gessha detailed a home setup that uses two Nvidia RTX 3090 GPUs in an i5-8600K workstation to host Qwen 3.8 27B and Qwen 3-VL 4B models via llama-server, alongside three mini PCs, a Lenovo Thinkpad T14 Gen 5, and a QNAP TS-231P NAS. gessha asked whether the distribution of services is reasonable and what expandable platform to use for moving the 3090s to a multi-GPU system, and a commenter recommended a used AMD Threadripper 3960X Proxmox build with 2x16 and 2x8 PCIe 4.0 slots, three NVMe slots, eight SATA ports, and built-in 10G networking.

read2 min views2 publishedSep 15, 2026
Homelab critique
Image: Forum (auto-discovered)

gessha 1

I have a somewhat unconventional way of heating my house, setup listed below.

workstation - i5-8600K, 32GB DDR4-2133, 2x Nvidia RTX 3090, Ubuntu 26.04 (friend’s e-waste aside from the 3090s) - Used for: llama-server hosting models (Qwen 3.8 27B, Qwen 3-VL 4B), Ubuntu 26.04 VM guest as LLM blast zone #

laptop - Lenovo Thinkpad T14 Gen 5, Ubuntu 26.04 - used for laptop purposes #

mini PC 1 - i5-4570T, 8GB DDR3-1600, integrated GPU, Debian 12 (decommissioned corporate mini PC) - jellyfin, couchdb(Obsidian), searxng, backups, custom webapp #

mini PC 2 - Ultra 5 235, 16GB DDR5-5600, integrated GPU, Debian 13 (questionably decommissioned corporate mini PC) - docker VM with docker containers: hister, openwebui, forgejo, grafana, postgres, label-studio, minio, etc #

mini PC 3 - (not set up yet) - ??? - I haven’t checked exactly what it is, probably similar to proxmox mini PC 1 (decommissioned corporate mini PC) #

NAS - QNAP TS-231P Context: I use the 3090s for coding via Pi+Qwen 3.8 27B Q4_K_M, Q8 KV and for custom apps where it’s used for structured data extraction from text, and images.

Some questions:

Is this a reasonable setup and distribution of services? #

Is there a better way of organizing the services over the compute? #

if I want to move the 3090s to a more expandable system with room for more GPUs, what would you recommend? I’ve previously thought about an old Threadripper for a 4-card setup but I’ve also oogled EPYC CPU+mobos I’m currently running a threadripper 3960x for my proxmox rig. It has been rock solid for the year I have had it. Got it used off of ebay.

Board has 2x 16 and 2x 8 pcie 4.0 slots. Three nvme slots and 8 sata ports with built in 10g networking.

Should give you plenty of room to grow the workstation.

Also, if you wanted to play with proxmox. You could put the workstation in a VM, and then run a bunch of your other services on the threadripper as well. That might free up a rig or two for you.

── more in #ai-infrastructure 4 stories · sorted by recency
── more on @gessha 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/homelab-critique] indexed:0 read:2min 2026-09-15 ·