The Most Powerful Mini PC with AMD Ryzen™ AI Max+ 395 for Running Local AI Models
If you're looking for a compact computer that
still packs exceptional processing power capable of running massive AI models
locally—without relying on cloud servers—the GMKtec EVO-X2 might be
exactly what you need. This mini PC is built around the latest AMD Ryzen™ AI
Max+ 395 processor, equipped with a second-generation XDNA 2 neural
processing unit (NPU), making it one of the first Windows 11 Pro mini PCs able
to run large language models with up to 70 billion parameters (70B) entirely
locally—no internet connection required, and no data ever leaves the device.
In this article, we'll break down the full
technical specifications, AI performance, design and cooling system,
connectivity options, and pricing for this device.
Key Highlights :
-
AMD Ryzen™ AI Max+ 395 processor with 16 cores
/ 32 threads, boosting up to 5.1 GHz.
-
Radeon 8060S GPU built on RDNA 3.5 architecture
with 40 compute units, delivering performance comparable to a dedicated RTX
4070.
-
NPU with a combined AI performance of up to 126
TOPS, up to 2.2x more power-efficient than an RTX 4090 when running LM Studio.
-
LPDDR5X unified memory running at 8000 MHz,
with configurations up to 128GB.
-
Storage expandable up to 16TB across two M.2
2280 slots over PCIe Gen4.
-
Advanced connectivity: WiFi 7, Bluetooth 5.4,
and a 2.5Gbps Ethernet port.
-
Triple-fan cooling system (Max 3.0) with 13
lighting patterns and 16 customizable cooling modes.
-
Windows 11 Pro pre-installed, with full
compatibility for Linux distributions such as Ubuntu.
Full Technical Specifications:
|
Component |
Details |
|
Processor (CPU) |
AMD
Ryzen™ AI Max+ 395 – 16 cores / 32 threads, TSMC 4nm process, up to 5.1 GHz,
16MB L2 cache and 64MB L3 cache, sustained TDP of 120W (peak up to 140W) |
|
Graphics (GPU) |
Radeon
8060S, RDNA 3.5 architecture, 40 compute units |
|
AI Processor (NPU) |
XDNA 2
architecture, 50 TOPS from the NPU alone, 126 TOPS combined across the whole
chip |
|
Memory (RAM) |
Unified
LPDDR5X (non-upgradable) at 8000 MHz, available in 64GB or 128GB
configurations |
|
Primary Storage |
PCIe 4.0
M.2 2280 SSD, 2TB or 4TB (does not support SATA drives) |
|
Storage Expansion |
Two
additional M.2 2280 slots, up to 16TB combined (8TB per slot via PCIe Gen4
x4) |
|
Video Outputs |
HDMI,
DisplayPort, and USB4, supporting resolutions up to 7680×4320 at 60Hz |
|
Wireless & Wired Networking |
WiFi 7
(RZ717/MT7925) and a 2.5Gbps RJ45 Ethernet port |
|
Bluetooth |
Bluetooth 5.4 |
|
Power Adapter |
DC 19.5V
/ 11.8A (approx. 230W), 100–240V AC input |
|
Front Panel Ports |
2x USB-A
3.2 Gen2, 3.5mm headphone jack, USB Type-C (USB4.0), SD4.0 card reader, power
button and fan/performance mode controls |
|
Rear Panel Ports |
DisplayPort
1.4 HBR3, HDMI 2.1, USB-A 3.2 Gen2, 2x USB-A 2.0, USB Type-C (USB4.0), 3.5mm
audio jack, DC power input, and 2.5Gbps RJ45 Ethernet port |
|
Operating System |
Windows
11 Pro (pre-installed), fully compatible with Linux and Ubuntu |
|
Dimensions & Build |
193 ×
185.8 × 77 mm, CNC-machined metal chassis for panels B and C, spray-coated
plastic for panel A |
AI
and Large Language Model (LLM) Performance :
AI performance is the standout strength of the
EVO-X2. Thanks to its large unified memory of up to 128GB, the device can run
very large language models entirely locally and with complete privacy, with no
need to send data to external servers.
|
Model |
Speed |
|
Deepseek-R1: 32B |
9.26 tokens/sec |
|
OpenAI GPT-OSS: 120B |
19.25 tokens/sec |
|
OpenAI GPT-OSS: 20B |
57.15 tokens/sec |
|
Qwen3: 30B |
55.02 tokens/sec |
|
Qwen3: 235B |
11 tokens/sec |
|
Llama4: 109B-Q4KM |
13.83 tokens/sec |
|
QWQ: 32B |
9.65 tokens/sec |
These figures come from benchmarks on the 128GB
variant, which also supports running massive models such as DeepSeek-R1 70B and
Llama 4 109B, while the 64GB variant is limited to models up to 32B.
Design and Cooling System :
The EVO-X2 features an elegant dual-layer,
two-tone design (black and silver), with a sturdy metal chassis that gives it a
professional look suited for workspaces and studios. The Max 3.0 cooling system
uses a 360-degree convection channel with three heat pipes and two
high-performance fans capable of handling peak power consumption of up to 140W
while keeping noise levels low. The device also offers 16 cooling modes and 13
customizable lighting patterns, controllable via the front-panel buttons.
Connectivity :
The device comes with a rich set of ports that
make it suitable for a complete workstation setup: quad-display support via
HDMI 2.1, DisplayPort 1.4, and two USB4 ports, plus a 2.5Gbps Ethernet port and
blazing-fast WiFi 7 (up to 46.4 Gbps theoretical speed)—making it an excellent
choice for gaming, design work, and running local AI servers all at once.
Versions and Pricing :
|
Version |
Price (EU Market) |
|
64GB RAM + 1TB SSD |
€1,949.99 (down from €2,299.99) |
|
96GB RAM + 2TB SSD |
€2,699.99 |
|
128GB RAM + 2TB SSD |
€3,199.99 |
The device comes with a full one-year warranty
from GMKtec.
Where
to Buy :
You can check the latest prices and available
deals, and order the device directly through the link below:
🛒 Get the GMKtec EVO-X2 Now
* This is an
affiliate link. We may earn a small commission if you make a purchase through
it, at no extra cost to you.
The GMKtec EVO-X2 is an exceptional choice for
anyone looking for a compact mini PC that delivers processing, graphics, and AI
performance rivaling mid-to-high-end desktop machines. Whether you're a
developer who wants to run large language models locally, a designer who needs
strong graphics performance, or a gamer looking for a small yet powerful
machine, this device is well worth serious consideration.
