What the EVO-X3 Is: A Mini PC AI Workstation for Local AI
The GMK EVO-X3 is a mini PC AI workstation that fits roughly into a PlayStation 4-sized chassis while offering a 16-core Ryzen AI processor, up to 128GB RAM, and expansion for external GPUs, targeting developers and creators who want portable AI computing without relying on the cloud. Built around AMD’s Ryzen AI Max+ 395 APU, it packs 16 cores and 32 threads with an 80MB cache, integrated Radeon 8060S graphics, and an NPU rated at 50 TOPS for on-device inference. GMK is pairing this with up to 128GB of LPDDR5X-8000 memory and 2TB or 4TB of PCIe 4.0 NVMe storage, turning the EVO-X3 into a 128GB RAM mini PC that behaves more like a compact workstation than a typical mini PC. A global launch is set for June 29, following early access registration on June 22.
Ryzen AI Max+ 395 and 128GB RAM: Why This Mini Box Matters
At the heart of the EVO-X3 is AMD’s Ryzen AI Max+ 395, a 16-core, 32-thread Strix Halo chip clocked up to 5.1 GHz with integrated Radeon 8060S graphics and a 50 TOPS NPU. According to Liliputing, the 395 supports “up to 128GB LPDDR5-8000” memory, which GMK is using to push the system far beyond typical mini PC configurations. High-bandwidth unified memory helps local AI inference, as both CPU, GPU, and NPU can share a large pool without PCIe bottlenecks. GMKtec also hints at a future Ryzen AI Max+ PRO 495 option with support for up to 192GB RAM and slightly higher CPU, GPU, and NPU clocks, aimed at even larger models. For developers working with multi-billion-parameter LLMs, this 128GB RAM mini PC starts to resemble a trimmed-down AI workstation node rather than a living-room mini PC.
OCuLink eGPU Expansion: Scaling Beyond Integrated Radeon Graphics
The most strategic addition in the EVO-X3 is its OCuLink eGPU expansion port, offering a PCIe Gen4 x4 link to an external GPU dock. Earlier GMK systems relied on USB4 for external graphics, but OCuLink cuts protocol overhead and delivers more reliable bandwidth for AI compute cards or high-end gaming GPUs. Wccftech notes that the new port “will allow users to run a discrete graphics card externally,” giving the EVO-X3 an upgrade path to RTX 40 or 50-series GPUs or other PCIe accelerators. For AI developers, that means starting with the integrated Radeon 8060S and scaling to multi-10 TFLOP external GPUs as workloads grow, without moving to a tower case. Combined with dual PCIe 4.0 M.2 slots and 140W APU headroom, OCuLink positions the EVO-X3 as a flexible, portable AI computing box rather than a fixed-spec mini PC.
Competing with DGX Spark and Enterprise AI Boxes in a Smaller Footprint
GMKtec is pitching the EVO-X3 as a credible alternative to larger AI workstations, including NVIDIA’s DGX Spark class of systems, but in a compact footprint. Wccftech describes the earlier EVO-X2, based on the same Ryzen AI Max+ 395, as “a strong competitor to the NVIDIA DGX Spark,” and the X3 builds on that with stronger I/O and cooling. Triple fans, a taller but slimmer case, and up to 140W power allow sustained performance closer to a desktop tower while occupying PS4-like desk space. GMK highlights use cases beyond hobbyist tinkering: local AI computing for LLMs, creative workloads such as 3D rendering and video production, and enterprise edge deployment. In that sense, the EVO-X3 tries to bridge mini PC and rack-style AI node, offering portable AI computing that can live on a developer’s desk or as a small edge box in an office.
Local-First AI Workflows and the Growing Compact AI Workstation Segment
Rather than a thin client for cloud services, the EVO-X3 is designed for local-first AI workflows. GMK bundles its Claw+Wrangler AI suite, promising 24/7 local agent and inference with “zero token costs and zero cloud dependency,” which aligns with developers who want predictable latency, better privacy, and offline capability. Gizmochina reports that GMK positions the 128GB configuration as capable of running models like a 235-billion-parameter Qwen3 entirely offline, especially when paired with OCuLink eGPU options. Connectivity such as WiFi 7, Bluetooth 5.4, USB4 with 40 Gbps data and 100W power input, HDMI 2.1, and 2.5GbE makes it easy to drop into an existing setup. With a June 29 launch date, the EVO-X3 underscores a wider shift: compact AI workstations are moving from niche mini PCs to serious tools for AI developers, creators, and edge deployments who want desktop-class performance in a portable box.
