- Ryzen AI Max+ 395128 GB unified memory
Full specifications
- Processor
- AMD Ryzen AI Max+ 395
- CPU cores
- 16 cores / 32 threads (Zen 5)
- Graphics
- AMD Radeon 8060S integrated
- NPU
- AMD XDNA 2, up to 50 TOPS
- Memory
- 128 GB LPDDR5X-8000 unified memory (soldered)
- Expansion
- PCIe x16 slot (x4 electrical), low-profile cards
- USB4
- 2 × USB4 v2, 80 Gb/s
- Networking
- 2 × 10 Gigabit Ethernet
- Power supply
- 320 W internal
- Wireless
- Wi-Fi 7
Good for
- Homelabs and small teams that want a 128 GB local-AI node with real expansion
- Rack deployment and multi-unit clusters
- Fast external storage over 80 Gb/s USB4 v2
Not ideal for
- Adding a full graphics card: the slot has four PCIe lanes
- Models under 30B, which run faster on a single 24–32 GB graphics card
- Buyers who want upgradeable memory
Minisforum
Minisforum MS-S1 Max
The MS-S1 Max has the same Ryzen AI Max+ 395 and 128 GB of unified memory, in a chassis designed for expansion. It adds a PCIe slot, two 80 Gb/s USB4 v2 ports and a built-in power supply, so it works as well in a rack or a homelab as it does on a desk.
£2,059Charged in euros at checkout: €2,469. Import duties and taxes included.In stock. Dispatched in 1 to 3 business days.
- Free shipping over €199, charged in euros
- Ships from our UK warehouse
- 30-day returns
- Import duties and taxes included
- Shipping, returns and warranty
The PCIe slot takes low-profile cards and runs at x4 lanes. It is ideal for a faster network card, not for adding a full-size graphics card.
Who it's for
Homelab builders who want a local-AI machine that can also serve the network and be clustered later.
Full specifications
- Processor
- AMD Ryzen AI Max+ 395
- CPU cores
- 16 cores / 32 threads (Zen 5)
- Graphics
- AMD Radeon 8060S integrated
- NPU
- AMD XDNA 2, up to 50 TOPS
- Memory
- 128 GB LPDDR5X-8000 unified memory (soldered)
- Expansion
- PCIe x16 slot (x4 electrical), low-profile cards
- USB4
- 2 × USB4 v2, 80 Gb/s
- Networking
- 2 × 10 Gigabit Ethernet
- Power supply
- 320 W internal
- Wireless
- Wi-Fi 7
About the Minisforum MS-S1 Max
The Minisforum MS-S1 Max pairs AMD's Ryzen AI Max+ 395 and 128 GB of unified memory with workstation features: a PCIe slot, two 80 Gb/s USB4 v2 ports, two 10 Gigabit Ethernet ports and a 320 W power supply built in. It runs the same large models as other 128 GB Ryzen AI Max machines. Buy it if you want a local-AI box that can also sit in a rack, take an expansion card and be clustered later.
Key features
- AMD Ryzen AI Max+ 395: 16 Zen 5 cores / 32 threads, up to 5.1 GHz
- 128 GB LPDDR5X-8000 unified memory on a 256-bit bus (soldered)
- Radeon 8060S graphics, 40 compute units at 2.9 GHz; up to 126 TOPS combined AI performance
- PCIe x16 slot, wired at PCIe 4.0 x4
- 2 × USB4 v2 (80 Gb/s) at the rear, 2 × USB4 (40 Gb/s) at the front
- 2 × 10 Gigabit Ethernet (RJ45), Wi-Fi 7, Bluetooth 5.4
- Built-in 320 W power supply; Performance, Balanced, Quiet and Rack modes
What it runs
We count about 75% of a unified pool as usable for models. On 128 GB that is roughly 96 GB, enough for 70B models at 8-bit and 120B-class models at 4-bit. A 120B model at 8-bit needs around 140 GB and does not fit on one unit. Check any model against this machine in our model fit tool.
| Model size | 4-bit | 8-bit |
|---|---|---|
| 7–8B | Fits | Fits |
| 13–14B | Fits | Fits |
| 30–34B | Fits | Fits |
| 70B | Fits | Fits |
| 120B+ | Fits | Does not fit |
Minisforum documents two-unit and four-unit clusters for models larger than one machine can hold. That works, but it is a project: plan the networking and the software split before you buy a second unit.
Performance in context
As with every Ryzen AI Max+ 395 system, memory capacity is the point, not bandwidth. Models that fit on a 24–32 GB graphics card run faster there; the MS-S1 Max is for the models that do not. The four power modes matter in practice: Quiet (60 W) suits a desk, Performance (130 W) suits long batch jobs, and Rack mode is meant for sustained load in a cabinet.
Ports and expansion
- PCIe x16 slot at PCIe 4.0 x4: suited to a network card, capture card or storage controller
- Two M.2 2280 slots: one PCIe 4.0 x4, one PCIe 4.0 x1, up to 8 TB each, RAID 0 and 1 supported
- Rear: 2 × USB4 v2 (80 Gb/s), 2 × USB 3.2 Gen 2, 2 × USB 2.0, HDMI 2.1 FRL (8K at 60 Hz or 4K at 120 Hz), 2 × 10 GbE
- Front: 2 × USB4 (40 Gb/s) with DisplayPort 2.0 Alt Mode, 1 × USB 3.2 Gen 2, audio jack, dual microphones
Put the operating system and the models you use most on the x4 slot; the x1 slot is fine for bulk storage and backups. The PCIe slot has four lanes, so treat it as an expansion slot for I/O rather than a place for a full graphics card.
Power, cooling and size
The 320 W supply is inside the case, so one mains cable is all it needs. Cooling uses a copper heat spreader, six heat pipes, phase-change thermal material and two turbine fans up to 3,600 RPM, holding 130 W under load with peaks to 160 W. The aluminium chassis measures 222 × 206 × 77 mm, weighs 2.8 kg, stands upright or lies flat in a 2U rack, and slides open for access to the drives.
Software
Minisforum names llama.cpp, Ollama and LM Studio as typical tools for this machine, and AMD lists Windows 11, Ubuntu and RHEL for the processor. It does not run CUDA; for a CUDA desk machine with 128 GB, see the NVIDIA DGX Spark.
What to pair it with
A 10 Gigabit switch, a fast NVMe drive for the x4 slot and, if you plan a cluster, a second unit of the same configuration. If you do not need the PCIe slot or rack mounting, the Beelink GTR9 Pro offers the same chip and memory in a smaller case. The Ryzen AI Max+ 395 comparison sets the two side by side.
Good for
- Homelabs and small teams that want a 128 GB local-AI node with real expansion
- Rack deployment and multi-unit clusters
- Fast external storage over 80 Gb/s USB4 v2
Not ideal for
- Adding a full graphics card: the slot has four PCIe lanes
- Models under 30B, which run faster on a single 24–32 GB graphics card
- Buyers who want upgradeable memory
What it can run
Up to 120B+ at 4-bit, 70B at 8-bit (approximate)
Based on 128 GB of unified memory, about 75% of which the GPU can use. See every machine by model size
Guides for this product
Mini PCs · 4 min read
Ryzen AI Max+ 395 Mini PCs Compared
Ryzen AI Max+ 395 mini PCs compared: Beelink GTR9 Pro or Minisforum MS-S1 Max, 128 GB of unified memory, ports, expansion and the models each runs.
Read the guide →Home server · 4 min read
Mini PC Home Server and NAS: A Quiet Build Guide
Build a quiet home server or NAS on a mini PC: an all-flash box, a refurbished business desktop or a 128 GB machine that also runs local AI.
Read the guide →Reviews
Rating elsewhere
Rating from the Minisforum store, checked Nov 17, 2026. Not reviews of orders placed with us.
What reviewers say
ServeTheHome
Their favourite Ryzen AI Max system so far, thanks to its features and expandability. It costs more than a bare-bones machine with the same chip.
“Minisforum did a great job overall with the MS-S1 Max”
Read the reviewNotebookcheckScore: 84%
Strong performance, a high-quality case and standout connectivity. Fan noise at idle cost it a higher score; a later BIOS update made it much quieter.
Read the review
Atomrig customer reviews
No customer reviews yet. Bought this from us? Tell us how it runs.
Related products

Beelink
Beelink GTR9 Pro
- 128 GB unified
- AMD Ryzen AI Max+ 395
Runs up to 120B+ at 4-bit
On Windows the GPU can be given up to 96 GB of the 128 GB. Check that the model you plan to run fits in that before you buy.
£3,639
Beelink
Beelink ME Mini
- 12 GB
- Intel N150, 4 cores / 4 threads, up to 3.6 GHz
Memory is soldered at 12 GB. That is plenty for storage, backups and a few containers, but not for running many virtual machines.
£179
Apple
Apple Mac mini (M4 Pro)
- 24 GB unified
- 16-core; 20-core option
Runs up to 13–14B at 4-bit
Memory cannot be upgraded after purchase. If local AI is the goal, choose 48 GB or 64 GB now. With 24 GB, models above about 14B parameters are a squeeze.
£1,169
NVIDIA
NVIDIA GeForce RTX 5090
- 32 GB
- NVIDIA Blackwell (GB202)
Runs up to 30–34B at 4-bit
32 GB runs 30B-class models comfortably at 4-bit, but a 70B model will not fit on one card. For 70B, plan for two cards or a large unified-memory system.
£1,669