Unified 128 GB mini PCs for local AI: what runs and what to buy
We earn from qualifying purchases via affiliate links. Disclosure.
A system with 128 GB of unified memory runs gpt-oss 120B, Qwen3.5 122B-A10B, Llama 3.3 70B with room. Complete-system route: A Strix Halo 128 GB mini PC. Build route: None: unified memory is soldered. It can join a cluster over 10 GbE or USB4.
Good, Better, Best: Unified 128 GB
Lowest 128 GB price we saw
GMKtec EVO-X2 128GB / 2TB
Price shown at Amazon
gpt-oss 120B (65 GB) and Qwen3.5 122B-A10B (81 GB) fit in the pool (our estimate). Ports not verified.
Sold at Amazon.
Check price at Amazon (opens in a new tab)
Verified ports for clustering
Minisforum MS-S1 MAX 128GB / 2TB (Ryzen AI Max+ 395)
$3,919 Minisforum · checked Oct 27, 2026, 12:40 PM CT
Caveat
Dual 10GbE, USB4 V2 80 Gbps, full-length PCIe x16 slot, built-in 320 W supply (store page).
Store data says estimated shipping late November. The 128GB / 4TB version was $4,199.
Check price at Minisforum (opens in a new tab)
Add a NAS for your model library
Minisforum N5 Pro AI NAS, barebone (no RAM or drives; 128 GB OS storage)
$899 Minisforum · checked Oct 27, 2026, 12:40 PM CT
Caveat
Keep big model files on one box on your network. Barebone: no RAM or drives included. Ships mid-November per the store.
Store data says estimated shipping mid-November for the HX PRO 375 version
Check price at Minisforum (opens in a new tab)Models that start at this tier
| Model | Memory needed | Format | Context | Smallest tier (our estimate) | Source |
|---|---|---|---|---|---|
 gpt-oss 120BOpenAI · Text | 65 GB | MXFP4 (as shipped) | 128K | Unified 128 GBUnified 128 GB | Ollama: gpt-oss tags |
 Qwen3.5 122B-A10B (MoE)Alibaba Qwen · Text | 81 GB | Q4_K_M | 256K | Unified 128 GBUnified 128 GB (needs most of it as GPU memory) | Ollama: qwen3.5 tags |
 Nemotron 3.5 Super 120B-A12BNVIDIA · Text | 90 GB | Q4_K_M (default tag 80 GB) | 1M | Unified 128 GBUnified 128 GB (needs most of it as GPU memory) | Ollama: nemotron-3.5-super tags |
Bigger tiers run everything smaller tiers do. Memory is the Ollama download size at the listed format; add roughly 1–4 GB for context and runtime (our estimate).
Can it run? Model pages for this tier
- Can the Ryzen AI Max+ 395 (128 GB) run Gemma 4 31B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Gemma 4 26B-A4B (MoE)? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.8 27B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.5 122B-A10B (MoE)? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Nemotron 3.5 Super 120B-A12B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.5 27B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.5 35B-A3B (MoE)? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Mistral Small 3.2 24B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run gpt-oss 120B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Llama 3.3 70B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run gpt-oss 20B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Phi-4 14B? Fits
- Can the Ryzen AI Max+ 395 (128 GB) run Llama 3.1 8B? Fits
Complete system
GMKtec EVO-X2 128GB / 2TB
Price shown at Amazon
Sold at Amazon.
Check price at Amazon (opens in a new tab) Minisforum MS-S1 MAX 128GB / 2TB (Ryzen AI Max+ 395)
$3,919 Minisforum · checked Oct 27, 2026, 12:40 PM CT
Caveat
Store data says estimated shipping late November. The 128GB / 4TB version was $4,199.
Check price at Minisforum (opens in a new tab)Build it yourself
There is no build route for this tier. Unified memory is soldered, so capacity is fixed at purchase. If you want a desktop instead, two GPUs whose VRAM adds up is the alternative; see the Llama 3.3 70B guide.
MS-S1 MAX vs EVO-X2
The MS-S1 MAX's store page lists dual 10GbE, USB4 V2 at 80 Gbps, a full-length PCIe x16 slot and a built-in 320 W supply. Minisforum also says that in its own testing, two clustered units ran a 235B-parameter model at Q4. That is the maker's claim; we have not checked it. We could not verify the EVO-X2's ports during this session. See the side-by-side table in the gpt-oss 120B guide.
FAQ
What can a Unified 128 GB system run?
gpt-oss 120B, Qwen3.5 122B-A10B, Llama 3.3 70B with room. Memory figures come from the Ollama library and maker pages; the tier calls are our estimates.
Can I build a Unified 128 GB machine myself?
Not as one box: unified memory is soldered to the board. The DIY alternative is a desktop with two GPUs whose VRAM adds up.
Do you recommend used GPUs?
No. FitMyModel lists new retail hardware with a warranty only, from Amazon, Newegg and Minisforum.
