
This thing is truly amazing. Can run models up to 200 billion parameters. (Most of mine are between 70-120b, but my 24b "devstral" model is very useful for simple coding.) For a "personal AI" it's perfect. I use it to assist with sensitive client matters that I do not want to "share" with public AI services. Yes, it is brittle, but it is not hard to perform factory-restore using Nvidia's image - they have clear instructions posted and AI can help you if you need it. Be sure to use Nvidia's tool to create the boot thumb-drive. The unique hardware and driver situation makes this computer incompatible with other distros or OSes.



- Easy setup, it was ready to use immediately - Good cooling, and super quiet - Sturdy feel and design

- Easy setup - Super quiet - Simple remote connection options - Preloaded Jupyterlab is convenient - Playbooks always getting added - NVIDIA ecosystem in a small package















I was debating between this GIGAIPC board and ASRock N100DC-ITX Mini-ITX. The ASRock is much faster but does not have the storage I need, plus too many bad reviews, Plus, it's a consumer grade board This mITX-6412A comes with 5 SATA and (1) M.2 SATA = 6 Storage. + Very Low TDP + 2 NIC + Silent + Mini-ITX + 2 PWM Fans + Industrial grade (made to run constantly) All these were exactly what I was looking for. I put it to work immediately, with used 8GB RAM ($17) + Coral Accelerator Cards (G650-04527-01) ($40) + (3) WD HDDs and (2) 2.5" SSD (I already had the drives) + INTEL SSDSCKKB240G8 240GB D3-S4510 M.2 2280 SSD ($25 used) running proxmox + Home Assistant + Frigate + OPNSense in a mix of VMs and containers. I know it is tight with 8GB RAM, but I'm planning to replace with 16GB later when price get less crazy than current market