What the GMK EVO-X3 Is and Why It Matters
The GMK EVO-X3 is a PS4-sized mini PC workstation built around AMD’s Ryzen AI Max+ 395 processor, up to 128GB of LPDDR5X memory, dual PCIe 4.0 storage and OCuLink eGPU expansion, designed to run demanding local AI computing, creative and enterprise workloads without relying on the cloud. GMKtec positions the system as a compact alternative to traditional tower workstations and even small enterprise AI nodes, with a footprint closer to a game console than a rackmount server. Early access registration begins June 22, with a global launch on June 29, and the company has already teased a future Ryzen AI Max+ PRO 495 variant supporting up to 192GB of memory. According to Liliputing, both EVO-X3 configurations are “roughly the size of a PlayStation 4,” making them larger than typical mini PCs but far smaller than full desktop towers.
Ryzen AI Max+ 395 at the Core: CPU, GPU and NPU
At the heart of the EVO-X3 is AMD’s Ryzen AI Max+ 395, a 16-core, 32-thread Strix Halo APU with Zen 5 CPU cores, an 80MB cache and integrated Radeon 8060S graphics based on RDNA 3.5. The chip boosts up to 5.1 GHz and includes an NPU rated at 50 TOPS for on-device AI inference, balancing CPU, GPU and dedicated AI acceleration in one package. GMKtec configures the system with up to 128GB of LPDDR5X-8000 memory, giving the APU unified access to a large high-bandwidth pool, useful for large language models and GPU workloads. Wccftech notes that EVO-X3 offers Balanced and Performance modes at 54W and 80W respectively, within a 140W system envelope, supported by triple-fan cooling. For users needing more headroom later, GMK has confirmed a Ryzen AI Max+ PRO 495 option with slightly higher clocks and support for up to 192GB memory.
128GB RAM Mini PC for Local AI Computing
The standout specification of the EVO-X3 is its 128GB RAM configuration, which pushes this mini PC firmly into workstation territory. GMKtec is targeting local AI computing scenarios such as running and testing large language models, multimodal agents and complex pipelines entirely offline. Gizmochina reports that GMKtec markets the 128GB configuration as capable of handling massive models, including examples like the 235 billion parameter Qwen3, on-device. Storage is equally workstation-like: two PCIe 4.0 x4 M.2 slots support 2TB or 4TB NVMe options, with user-replaceable drives for expansion. RAM is soldered, so capacity must be chosen at purchase, but the large unified memory pool benefits both CPU and integrated GPU workloads. This combination of high memory bandwidth, generous capacity and fast storage makes the EVO-X3 attractive for developers, researchers and creative professionals who want AI and data workloads close to their files instead of in the cloud.
OCuLink eGPU Expansion and I/O for Scalable Performance
The EVO-X3’s most strategic upgrade over the EVO-X2 is its OCuLink port, which provides a PCIe Gen4 x4 link for external graphics docks. This gives the mini PC workstation a clear path to stronger GPU acceleration by connecting desktop-class RTX 40- or 50-series cards, a faster and cleaner solution than USB4-based eGPU enclosures. Wccftech highlights the OCuLink addition as the “first and most important upgrade,” enabling far higher external GPU performance than before. The system keeps full-function USB4 ports (40 Gbps with video and 100W power input), HDMI 2.1, WiFi 7, Bluetooth 5.4 and 2.5GbE LAN, plus two PCIe 4.0 M.2 slots. GMK’s triple-fan, triple-heat-pipe cooling and up to 140W power budget give enough thermal headroom for sustained CPU and NPU loads, while OCuLink lets users scale GPU compute over time instead of replacing the entire system.
Challenging Traditional AI Workstations and Enterprise Nodes
GMKtec is pitching the EVO-X3 as a compact alternative to enterprise AI boxes like NVIDIA’s DGX Spark and to full-size desktop workstations. Compared with tower systems, the PS4-sized chassis and vertical design reduce desk footprint while keeping enough volume for serious cooling and I/O. According to Wccftech, the EVO-X3 “is going to be a powerful workstation PC for local AI,” bridging personal and professional use cases with configurations up to 128GB memory and 4TB storage. GMK’s Claw+Wrangler AI suite adds 24/7 local agent and inference tools, aiming for “zero token costs and zero cloud dependency.” That combination of on-device NPU, integrated Radeon graphics, OCuLink eGPU expansion and rich connectivity makes the EVO-X3 appealing to creators, developers and edge deployments that need private, low-latency AI without renting remote GPUs or managing large racks of servers.
