What Compact AI Workstations Are and Why They Matter
Compact AI workstations are small-form-factor computers that fit in console-sized cases yet deliver desktop-class CPUs, integrated GPUs, and NPUs tuned for local AI workloads, enabling tasks like large language model inference, content creation, and data analysis without relying on cloud servers. The new wave of Ryzen AI Max+ mini PCs, led by systems such as Gmktec’s EVO-X3, pushes this idea further by combining 16-core processors, high-bandwidth LPDDR5X memory, and advanced connectivity in a chassis roughly the size of a PlayStation 4. This form factor blurs the line between mobile and stationary computing: small enough to move between home, office, or studio, but powerful enough to function as a primary machine. For developers, researchers, and creative workers, it signals a shift from bulky towers or remote clusters to portable AI computing boxes that live on the desk, not in a rack.
Inside the EVO-X3: Ryzen AI Max+ and 128GB in a Mini PC
Gmktec’s EVO-X3 centers on AMD’s Ryzen AI Max+ 395, a 4 nm, 16-core/32-thread processor with Radeon 8060S graphics and an XDNA2 NPU rated at 50 TOPS of dedicated performance. Liliputing notes that the Ryzen AI Max+ 395 configuration supports “up to 128GB LPDDR5-8000,” while a planned Ryzen AI Max+ PRO 495 option will raise the ceiling to 192GB, still in a case about the size of a PS4. Gizmochina reports that the current EVO-X3 ships with 128GB of LPDDR5X memory at 8000 MT/s and can assign up to 96GB of that as VRAM for local models. Altogether, the system is rated for up to 126 TOPS of total AI compute, aligning it with high-end desktops. Dual PCIe 4.0 x4 M.2 slots allow up to 16TB of storage, and Wi-Fi 7 plus USB4 ports keep this 128GB mini PC connected at workstation speeds.
OCuLink eGPU Support and Modular Performance
A key difference between these compact AI workstations and traditional mini PCs is native OCuLink eGPU support. Both Liliputing and Gizmochina highlight that the EVO-X3 includes an OCuLink port wired to PCIe 4.0, enabling direct attachment of external graphics docks. According to Gizmochina, the system can connect to desktop GPUs including Nvidia’s RTX 40-series and 50-series, giving users a clear upgrade path if integrated Radeon 8060S graphics are not enough. For AI practitioners, that means starting with the integrated GPU and 96GB of shared VRAM, then adding a desktop-class card later for heavier vision or CUDA-centric workloads. Because OCuLink avoids some of the bandwidth and latency penalties of generic USB enclosures, this design keeps portable AI computing practical: the core Ryzen AI Max+ mini PC travels easily, while the external GPU can stay on a desk where it belongs.
Local AI Inference and Personal AI Services on the Desk
The EVO-X3’s specification is clearly aimed at local AI inference and personal AI services. Gizmochina notes that Gmktec allows “users to allocate up to 96GB of this system memory directly as VRAM” and claims the machine can run large local models like Qwen3 235B, with LM Studio performance that can surpass a desktop RTX 4090 in some workloads. The pre-installed Claw AI software, tuned for AMD’s ROCm ecosystem, reduces setup time for large language models and other AI tools. With up to 126 TOPS of AI compute on tap, these compact AI workstations are built to host chatbots, code assistants, and media generators locally, reducing dependence on cloud APIs. That shift has privacy and latency benefits: sensitive data never leaves the machine, and responses depend on local resources rather than network conditions or shared cloud capacity.
Bridging Mobile and Desktop: A New Class of Workstation
Physically, the EVO-X3 sits between a gaming console and a classic tower, using a CNC-machined metal chassis with three copper heat pipes and three fans to keep sustained workloads in check. Power profiles for silent, balanced, and performance modes make it easier to tune noise and thermals without giving up capability. This balance shows how compact AI workstations are redefining what a “desktop” can be: a PS4-sized box that still supports dual 8K outputs, 2.5G Ethernet, Wi-Fi 7, and full-featured USB4. Other vendors are joining in; Gizmochina notes Peladn’s HO5 mini PC with a Ryzen AI 9 HX 470 and OCuLink eGPU expansion, suggesting a broader move toward high-memory, AI-focused mini PCs. As these systems spread, the default assumption that serious AI workloads require either cloud clusters or oversized towers looks increasingly outdated.
