01-04-2026, 03:19 PM
You have a very solid build. It is efficient, quiet, expandable and perfect for a long lasting TrueNAS system that can handle VMs and local AI work. I would not change the motherboard and only upgrade NVMe capacity if you want more room for models and VMs.
Jonsbo N6
Great case for airflow and a clean layout. Plenty of room for drives and a full GPU, so this works very well for a mixed NAS plus AI box.
ASRock Rack W680D4U
This is one of the best TrueNAS friendly boards you can buy. You get IPMI, reliable Intel NICs, ECC support, and 10G built in. For a home server that needs to be stable for years, this is the right type of board. You are not likely to find a “better” option unless you jump to server grade EPYC or Xeon E5 gear, which brings higher power draw and much more noise.
Intel i5 14500
Excellent choice. It gives you plenty of cores for VMs and handles TrueNAS scale’s Kubernetes apps easily. It also supports ECC on W680, so no issues here.
2 x 32GB DDR5 ECC
Perfect. Sixty four GB is a sweet spot for VM workloads and LLM caching.
2 x 1TB NVMe
Use these as mirrored boot or VM storage. For TrueNAS Scale, you do not need ultra high endurance drives unless you plan on heavy database workloads.
4 x 10TB NAS drives
A nice size for a RAIDZ1 or RAIDZ2 pool. For long term reliability and because you are running VMs and LLM models, I would lean toward RAIDZ2.
RTX 4060 Ti
Totally fine for local LLM inference. Models like Llama 13B and 70B quantized versions will run well.
850W PSU
More than enough and leaves room for future GPU upgrades.
Noctua NH-U12S
No objections. Quiet and fits nicely in the N6.
Would I change anything?
The motherboard is already one of the strongest choices for this purpose. The only alternatives with similar features are:
Supermicro X13SAE or X13SAE-F
Comparable but more workstation oriented. You would lose some of the server grade feel the ASRock Rack board provides.
Supermicro server boards with SP6 or Xeon W series
Better for lots of PCIe lanes, but far more expensive and not necessary for your use case.
So unless you have a special requirement, your chosen board is ideal.
A small optional improvement
If your budget allows, consider using 2 x 2TB NVMe instead of 1TB. VM disks and AI model caching tend to grow very quickly.
Jonsbo N6
Great case for airflow and a clean layout. Plenty of room for drives and a full GPU, so this works very well for a mixed NAS plus AI box.
ASRock Rack W680D4U
This is one of the best TrueNAS friendly boards you can buy. You get IPMI, reliable Intel NICs, ECC support, and 10G built in. For a home server that needs to be stable for years, this is the right type of board. You are not likely to find a “better” option unless you jump to server grade EPYC or Xeon E5 gear, which brings higher power draw and much more noise.
Intel i5 14500
Excellent choice. It gives you plenty of cores for VMs and handles TrueNAS scale’s Kubernetes apps easily. It also supports ECC on W680, so no issues here.
2 x 32GB DDR5 ECC
Perfect. Sixty four GB is a sweet spot for VM workloads and LLM caching.
2 x 1TB NVMe
Use these as mirrored boot or VM storage. For TrueNAS Scale, you do not need ultra high endurance drives unless you plan on heavy database workloads.
4 x 10TB NAS drives
A nice size for a RAIDZ1 or RAIDZ2 pool. For long term reliability and because you are running VMs and LLM models, I would lean toward RAIDZ2.
RTX 4060 Ti
Totally fine for local LLM inference. Models like Llama 13B and 70B quantized versions will run well.
850W PSU
More than enough and leaves room for future GPU upgrades.
Noctua NH-U12S
No objections. Quiet and fits nicely in the N6.
Would I change anything?
The motherboard is already one of the strongest choices for this purpose. The only alternatives with similar features are:
Supermicro X13SAE or X13SAE-F
Comparable but more workstation oriented. You would lose some of the server grade feel the ASRock Rack board provides.
Supermicro server boards with SP6 or Xeon W series
Better for lots of PCIe lanes, but far more expensive and not necessary for your use case.
So unless you have a special requirement, your chosen board is ideal.
A small optional improvement
If your budget allows, consider using 2 x 2TB NVMe instead of 1TB. VM disks and AI model caching tend to grow very quickly.