Custom NAS build

[Enquiries]

Good evening NAS Compares forum!

I am writing this message regarding a custom NAS build I would like to construct from my home as an upgrade from my Ugreen DPX2800. While the Ugreen NAS has been an amazing bit of kit I use to host my applications such as Plex, Home assistant & Jellyfin. I want to take things up a level.

Having sold off my old PC, I want to create a NAS build that can run some of the top of the line LLM's fully locally on it's system, encode Plex at 4K quality easily and store my personal files. I want to stress that my energy company provides me with a fixed power bill so feel free to stress that wattage! I currently have three 3.5 HDD's, and are planning to buy more.

I don't really have any specific requirements other than to build a beast of a system! Perhaps picking out a specialised PC case for this build?

Thank you :]

[ed]

Hi, thanks for the message. If you’re upgrading from the UGREEN DXP2800 and want a true powerhouse NAS that can handle local LLMs, 4K Plex encoding, and large-scale storage, then you’re looking at a workstation-grade build rather than a typical NAS. Given your fixed power bill, you can afford to go big. Here’s what I’d suggest for a custom build that balances performance, expandability, and reliability.

Case:
Go for a Jonsbo N3 or Fractal Design Node 804. Both can hold 8–12 drives comfortably, have excellent airflow, and look clean enough for home or office setups.

CPU:
Choose the AMD Ryzen 9 7950X or Ryzen 9 7950X3D if you want maximum multi-core performance for AI workloads. Alternatively, the Ryzen 7 7800X3D gives you incredible efficiency and plenty of power for LLM inference and transcoding.

Motherboard:
Use an X670E or B650E board with at least 6 SATA ports, dual M.2 slots, and a full PCIe x16 slot for GPU expansion.

Memory:
64 GB DDR5 (expandable to 128 GB if you plan to load larger models or multiple containers).

GPU (optional but recommended):
An NVIDIA RTX 4070 Ti SUPER or RTX 4080 gives you both CUDA cores and VRAM for LLMs and Plex hardware transcoding. You can also experiment with quantized models locally using Ollama or LM Studio.

Storage:
• 1 TB NVMe SSD for system and apps
• 2 TB NVMe SSD for AI cache and Docker containers
• 6–8× large HDDs (20 TB+ each) for main storage (WD Red Pro or Seagate Exos)

PSU:
850–1000W 80+ Gold unit from Seasonic or Corsair, modular and quiet.

OS Options:
Run Proxmox VE as your hypervisor. Use VMs or containers for TrueNAS SCALE (for file storage), Plex, Home Assistant, and any LLM stack like Ollama, Text Generation WebUI, or LM Studio.

This build will give you desktop-grade AI performance, handle 4K Plex encoding with ease, and give you space for over 120 TB of storage. It’s a serious machine that doubles as a NAS and AI lab.