A desktop-class AI mini PC in a shoebox footprint
The GMKtec EVO-X3 is a premium AI mini PC that compresses desktop-class CPU, GPU and neural processing performance into a compact workstation aimed at creative professionals and AI developers who want serious local compute power without a full tower on their desk.
The main story is simple: GMKtec is betting that some users will trade cost for density. Following an initial limited release, the EVO-X3 is now being launched globally as an AI workstation with a clear pitch—run demanding large language models and content creation workloads locally, in a chassis that looks more like a console than a traditional tower. Inside that metal shell sits AMD’s Ryzen AI Max+ 395, a 16-core, 32-thread Zen 5 chip with boost clocks up to 5.1GHz, paired with an XDNA 2 NPU that pushes total system AI compute to 126 TOPS. In other words, this is not a cute hobby box; it is a serious creative professional PC that happens to be small.
Ryzen AI Max+ 395: 126 TOPS performance in a compact workstation
At the heart of the EVO-X3, the Ryzen AI Max+ 395 turns the mini PC into a specialized AI machine rather than a generic small form factor system. The 16-core/32-thread CPU configuration gives it the multi-threaded muscle you’d normally expect from a high-end desktop, while the integrated Radeon 8060S iGPU with 40 RDNA 3.5 compute units fills the role of a workstation-class graphics engine. On top of that, the XDNA 2 NPU delivers 50 TOPS by itself, contributing to a total of up to 126 TOPS of AI power for local inference.
This matters because it shifts what a compact workstation can do. 126 TOPS performance is enough to run sizeable large language models and complex AI agents without shipping data to the cloud, a clear value proposition for developers working with sensitive datasets or iterative experimentation workflows. One quotable way to frame it is: “Ryzen AI Max+ 395 as a whole brings up to 126 TOPS of AI power, which is enough for intensive local AI workloads such as running large language models.” For anyone building or testing AI mini PC deployments, this is close to a plug-and-play reference box.
Unified memory and Radeon 8060S: a creative professional PC without a discrete GPU
GMKtec’s configuration choices make it clear that the EVO-X3 is built as much for artists and engineers as for pure AI researchers. Every configuration ships with 128GB of LPDDR5X memory at 8000 MT/s in a unified architecture, and up to 96GB of that RAM can be assigned directly to the GPU. That is a clever answer to a familiar problem: most AI and 3D workloads are bottlenecked by VRAM, and this approach gives the integrated Radeon 8060S more headroom than many midrange discrete cards.
The Radeon 8060S is described as “the most powerful integrated graphics available in the mainstream market,” and is positioned as sufficient for 3D rendering, gaming, and 4K video editing while the NPU handles AI tasks. For video editors and 3D artists who do not want to deal with external GPUs, this makes the EVO-X3 a credible all-in-one creative professional PC. It will not replace a dual-GPU render node, but for many real-world timelines and preview tasks, the balance of CPU threads, iGPU capability and NPU acceleration is more than enough.
Storage, expandability and thermals: thoughtful, but not cheap
Storage and connectivity decisions show that GMKtec expects this box to sit at the center of a serious workflow. Out of the box, buyers can choose between 2TB or 4TB PCIe 4.0 NVMe SSD configurations, with two M.2 2280 slots supporting up to 16TB if you upgrade later. There is USB4 with 100W power delivery and 4K output, HDMI 2.1, 2.5G Ethernet, Wi‑Fi 7, Bluetooth 5.4, and an OCuLink port for external GPUs operating in PCIe 4.0 x4 mode. You can plug in an RTX 40- or 50‑series card if you need more grunt, even if that x4 link will cap peak performance.
Running this kind of hardware in a slim mini chassis is non-trivial, so GMKtec provides three power modes—54W silent, 85W balanced, and 120W performance with peaks to 140W—cooled by dual fans and three heat pipes. That should keep sustained AI or render workloads stable, though anyone expecting silence under full load is kidding themselves. Overall, the design is cleverly engineered, but everything from the OCuLink dependence to the thermal envelope reminds you that this is still a compact workstation, not magic.
A $3,600 AI workstation: who is the EVO-X3 really for?
All of this performance comes at a clear price. The base configuration with 128GB of memory and a 2TB SSD costs USD 3,600 (approx. RM16,800), while the 4TB version is USD 3,849 (approx. RM17,960). That instantly puts the EVO-X3 in premium territory; you could build a full-size tower with a high-end CPU and a serious discrete GPU for similar money. But that misses the point of this product. It targets people who see value in density, portability and local privacy: independent AI developers, small studios, and creative nomads who want a single, compact box that can sit on a desk, be carried between offices, or even double as a home lab node.
GMKtec is rolling this out globally starting July 6, 2026, in 128GB+2TB and 128GB+4TB configurations. That timing matters: AI tools are maturing, and more professionals are shifting from cloud-only to hybrid setups. In that context, the EVO-X3 looks less like an indulgence and more like a specialized instrument. It will not be the right choice for budget-conscious users or gamers chasing pure frames per dollar. But if you care more about 126 TOPS performance, unified 128GB memory and a compact footprint than about raw price, this AI mini PC is one of the few ready-made answers on the market.
