Best Home AI Server: questions & answers

Our editorial choice is Olares One. It is quiet by Olares' lab figures, has 24 GB of VRAM with CUDA and gives each family member an account. A used RTX 3090 tower costs less if you build it, and a Mac Studio holds larger models.

From nothing, if you reuse a PC, to 4,999 USD for Olares One (Olares shop, 13 November 2026). In between, Apple lists a 64 GB Mac mini at 3,199 USD, and SpecPicks estimates a used RTX 3090 tower at about 3,800 USD without a case.

Olares says Olares One measured 19 dB at idle and under 39 dB at maximum load in its lab. A tower with a desktop GPU makes more heat, so read reviews of the exact build before putting it in a living room.

Yes. Olares OS has separate accounts with roles. In Olares' own test, the box served 8 requests of Qwen3-30B-A3B at once at about 81 tokens per second each.

According to Olares' FAQ, Olares One can mount a NAS share over SMB and run apps like Immich on the photos and videos stored there. It can also back up to a NAS over SMB.

Olares lists 30 W in standby, which is 0.72 kWh a day. Multiply by your electricity rate, then add the hours of heavy use.

Yes. Olares includes a VPN and remote access through the LarePass app, without a static IP. On a DIY server, Tailscale does the same job.

On 24 GB of VRAM, models up to about 32B at 4-bit, which need roughly 19-20 GB (approximate). A 70B model needs about 40 GB, which calls for a large-memory Mac Studio or a DGX Spark.