What the GMK EVO-X3 Is and Why It Matters
The GMK EVO-X3 is a compact desktop AI workstation that integrates a 16-core Ryzen AI Max+ 395 processor, up to 128GB of LPDDR5X memory, and dual NVMe storage into a PS4-sized chassis, targeting users who want enterprise-grade local AI performance without a full-size tower or reliance on the cloud. Instead of chasing gaming aesthetics, GMK positions the EVO-X3 as a mini PC AI workstation tuned for local inference, development, and creative workloads. According to Liliputing, the system will open early access registration on June 22, with a global launch set for June 29. That timing underlines GMK’s plan to ride the growing demand for compact desktop AI systems that can run sizeable language models, vector databases, and media pipelines directly on the desk, while leaving an upgrade path for external GPUs through high-speed OCuLink connectivity.
Ryzen AI Max+ 395 and the 128GB Mini PC Architecture
At the heart of the EVO-X3 sits AMD’s Ryzen AI Max+ 395, a 16-core, 32-thread Strix Halo APU built on Zen 5 with Radeon 8060S integrated graphics and a 50 TOPS NPU. GMK pairs it with up to 128GB of LPDDR5X running at 8000 MT/s and two PCIe 4.0 x4 M.2 slots for 2TB or 4TB configurations, creating a 128GB mini PC that blurs the line between mini PC and workstation. Wccftech notes that GMK is already planning a Ryzen AI Max+ PRO 495 version that will support up to 192GB of memory for even larger models. Thermal design is equally important: a triple-fan, triple heat-pipe cooling system and up to 140W power budget allow Balanced and Performance modes at 54W and 80W, which helps sustain clock speeds during long AI training runs or multi-hour renders in a compact desktop AI form factor.
OCuLink, USB4 and the eGPU Angle vs Traditional Desktops
A key differentiator for the EVO-X3 is its OCuLink (PCIe Gen4 x4) port, which gives the mini PC AI workstation a desktop-like expansion path without an internal PCIe slot. OCuLink provides a direct PCIe link for an external RTX 40- or 50-series GPU dock, offering lower latency and higher throughput than eGPU setups limited to USB4. GMK keeps USB4 on board too, with 40 Gbps data, video output, and 100W power input, alongside HDMI 2.1, WiFi 7, Bluetooth 5.4, and 2.5GbE. In practice, this means users can start with integrated Radeon 8060S graphics and later add a desktop-class GPU over OCuLink if their workloads grow. That modular path challenges the need for heavy ATX towers in many AI and creative studios, especially where desk space and acoustics favor a slimmer, PS4-like node over a full-blown workstation chassis.
Target Users: From Local AI Labs to Creative Studios
GMK is aiming the EVO-X3 at developers, creative professionals, and edge deployments that want portable AI computing without constant cloud access. The bundled Claw+Wrangler AI suite focuses on 24/7 local agents and offline inference, and GMKtec claims the 128GB configuration can run very large language models entirely offline. The company highlights three core use cases: local AI computing for LLM development and testing with lower latency and better privacy, creative workloads such as 3D rendering and video production that benefit from high memory bandwidth, and enterprise edge roles in education, healthcare, and government infrastructure. With its PS4-sized profile and vertical, slim design, the EVO-X3 fits as a compact desktop AI node on a shared bench, in a lab rack, or in a small studio, while still offering enough I/O to anchor multiple displays, storage arrays, and an OCuLink GPU dock.
Competing with Enterprise AI Workstations and DGX Spark
Strategically, the EVO-X3 pushes GMK further into the enterprise AI workstation space, but at a smaller and more flexible scale. Wccftech frames the Ryzen AI Max+ 395-based EVO-X line as a competitor to NVIDIA’s DGX Spark, especially for users who do not need massive multi-GPU clusters but still want serious local compute. Instead of a monolithic rack node, EVO-X3 offers a compact desktop AI experience that can be scaled horizontally—multiple units on a desk or edge rack—or vertically by adding OCuLink GPUs. Claw+Wrangler’s zero-token, zero-cloud dependency pitch will appeal to organizations with strict data controls or intermittent connectivity. While GMK has not disclosed pricing, the feature set suggests a niche between hobbyist mini PCs and full enterprise servers, giving small teams a way to experiment with on-prem AI infrastructure before committing to larger, more complex deployments.
