Hardware

Mini PCs & unified memory for local AI

Machines where the GPU shares system memory. Slower per token than a big GPU, but they can hold far larger models in a small, quiet box.

Models that run well on this tier

gpt-oss:120b

117B MoE (5.1B active) · MXFP4 · ~66.5 GB

Needs ~80GB of VRAM or unified memory (e.g. 96-128GB machines).

qwen3.6:35b

35B MoE (3B active) · Q4_K_M · ~25.5 GB

MoE flagship; needs 24GB+ or VRAM + RAM offload.

gemma4:26b

26B MoE (~4B active) · Q4_K_M · ~19.5 GB

Mixture-of-experts: big-model quality, small-model speed.

qwen3-coder:30b

30B MoE (3B active) · Q4_K_M · ~20.5 GB

Agentic coding MoE; fast even with partial offload.

llama3.3:70b

70B · Q4_K_M · ~44.5 GB

Needs 48GB+ VRAM or lots of unified memory.

Check your exact setup in the GPU picker → · Read the full guide →

Hardware picks

Mini PC with AMD Ryzen AI Max+ 395 (64-128GB)

Up to 128GB shared · bandwidth ~256 GB/s

Listed as ROCm-supported by Ollama. With 128GB it can load gpt-oss 120B or 70B models; MoE models run best.

Check price

Apple Mac mini (M4 Pro, 48-64GB)

Unified memory · bandwidth 273 GB/s

Silent, efficient and simple: Ollama and LM Studio use the Apple GPU via Metal/MLX out of the box.

Check price

Apple Mac Studio (64GB+)

Unified memory · bandwidth Higher on Max/Ultra chips

The premium option for very large models. Check the memory bandwidth of the exact chip before buying.

Check price
Our advice: Choose unified memory for capacity (big models, quiet desk), and a discrete GPU for speed. Mixture-of-experts models like Qwen3.6 35B-A3B shine on these machines because only ~3B parameters are active per token.

We don't list prices because they change daily. Bandwidth figures are manufacturer specs. Disclosure: "Check price" links may be affiliate links (rel="sponsored"); we may earn a commission at no cost to you.

Get the Local AI weekly

New models worth running, what fits which GPU, and setup fixes. One short email a week, free. Unsubscribe any time.

Lila Forge Kits

Local-LLM App Starter

A tested starter for building your own app on Ollama: chat, RAG and a GPU-fit checker. Skip the setup hours.

kits.lilaforge.com → VibeEngine

Best local LLM tools, ranked

Ollama, LM Studio, Open WebUI and llama.cpp scored from public evidence, with top picks.

vibeengine.lilaforge.com →