On September 6, 2026, GMKtec announced the EVO-X5 Pro, a flagship desktop AI computing platform powered by a cutting-edge processor and up to 192GB of unified memory. Designed to support offline local execution of large language models with up to 300 billion parameters, the system aims to lead the industry into the next era of agentic PCs.

Official Unveiling of the EVO-X5 Pro at IFA 2026

At the IFA 2026 exhibition in Germany, GMKtec officially announced its flagship mini PC, the EVO-X5 Pro, on September 6, 2026. This pioneering system is powered by AMD’s latest Ryzen AI Max+ PRO 495 processor and supports up to 192GB of unified LPDDR5X memory. By squeezing such immense memory capacity into an ultra-compact chassis, GMKtec has completely redefined what a mini PC can achieve. The system integrates advanced CPU, GPU, and NPU computing capabilities into a single desktop platform, allowing highly sophisticated AI workloads to run in a fully local and offline environment without ever sending data to external cloud servers.

In the era of corporate digital transformation, preventing data leaks and securing sensitive information have become critical priorities. This security focus is mirroring regulatory changes, such as the Japanese Ministry of Internal Affairs and Communications’ directive in April 2026 that restricts local government IT procurement strictly to government-certified devices. In this landscape, the EVO-X5 Pro serves as an ideal solution for public municipalities and enterprise development environments that require uncompromising data security while utilizing the latest artificial intelligence. Refer to the diagram below.

Empowering 300-Billion-Parameter Local Inference

The unified LPDDR5X memory featured in the EVO-X5 Pro operates at a blinding speed of 8533 MT/s, delivering an exceptionally wide memory bandwidth of approximately 273 GB/s. This wide-bandwidth, high-capacity memory pool significantly increases the upper boundary of the model sizes that can be loaded into memory. In GMKtec’s optimized environment, mixture-of-experts (MoE) large language models boasting up to 300 billion (300B) parameters can reside entirely in local memory to run ultra-low-latency local inference.

On older 128GB systems, local AI execution was capped at 70B-class models with 4-bit quantization, or MoE models up to roughly 200B parameters. Expanding the memory capacity by 50% to 192GB, however, brings complex 120B-class quantized models and massive 300B MoE models within practical local execution range. This transition allows professionals to execute resource-intensive tasks—such as deep document summarization, complex multi-file code generation, and secure internal business consulting—completely free of charge and with absolute privacy, bypassing the recurring fees and security risks of premium cloud AI subscriptions.

Hardware Architecture and Technical Specifications

The Synergy of 16-Core Zen 5, Radeon 8065S, and XDNA 2

The powerhouse behind the EVO-X5 Pro is the Ryzen AI Max+ PRO 495, a flagship APU engineered by AMD for local AI development and intensive professional workloads. The CPU portion leverages the cutting-edge Zen 5 architecture, housing 16 cores and 32 threads with a boost clock of up to 5.2 GHz. This ensures top-tier multi-threaded performance for data preprocessing, compilation, and general computing tasks essential for local AI pipelines.

On the graphics side, the APU integrates the Radeon 8065S, a powerful integrated GPU built on the RDNA 3.5 architecture. Equipped with 40 compute units (CUs) and a maximum clock speed of 3000 MHz (3.0 GHz), this iGPU supports hardware-accelerated ray tracing. Its rendering capabilities rival mid-range discrete mobile graphics, such as the RTX 4070 Laptop GPU, and desktop-class RTX 4060 GPUs. For specialized neural processing, AMD’s integrated XDNA 2 NPU delivers up to 55 TOPS of dedicated AI compute. This aggregate processing power efficiently handles background AI tasks and continuous productivity assistants with minimal power consumption.

160GB VRAM Allocation and Advanced Thermal Engineering

The unified memory architecture of the EVO-X5 Pro allows the CPU and GPU to share a single physical memory pool, removing the latency penalties typically caused by copying data between system RAM and discrete graphics VRAM. Remarkably, the system can allocate up to 160GB of its 192GB LPDDR5X memory directly to the GPU as VRAM. This design brings massive VRAM capacities—previously only accessible through expensive, multi-GPU workstation setups or enterprise-class data center nodes—straight to a palm-sized form factor.

To sustain this heavy hardware load over extended periods, GMKtec has fitted the EVO-X5 Pro with an exceptionally robust cooling system. The chassis incorporates a highly efficient, large-scale vapor chamber paired with three meticulously arranged silent cooling fans. Users can toggle between three performance modes designed for silent operation, balanced workloads, and extreme computation. This advanced thermal engineering prevents thermal throttling, maintaining peak processing performance even during around-the-clock local LLM inference, 3D rendering, or heavy gaming workloads. Refer to the diagram below.

Enterprise-Grade Reliability and the Road Ahead

24/7 Operational Integrity and Remote Administration

Beyond its raw computational performance, the EVO-X5 Pro is built for professional, enterprise-grade deployment. GMKtec has subjected this mini PC to rigorous quality-control benchmarks simulating continuous 24/7/365 operations, guaranteeing the durability needed to run the system as a dedicated local AI workstation or micro-server.

To streamline integration into corporate networks, the EVO-X5 Pro supports AMD DASH, an industry-standard out-of-band management technology. With a dedicated RJ45 DASH network port located on the rear panel, system administrators can securely monitor system health, deploy OS updates, and modify BIOS configurations remotely. Furthermore, a hardware-based dTPM 2.0 security chip is soldered onto the motherboard, protecting the system from unauthorized firmware manipulation or physical intrusion. This robust security makes the EVO-X5 Pro uniquely suited for processing highly sensitive data locally, such as patient medical records in hospitals or confidential litigation documents in law firms.

Driving the Agentic PC Era and Shaping the Market

As we move through the latter half of 2026, the artificial intelligence landscape is rapidly transitioning from simple query-and-response interfaces to agentic PCs capable of planning and executing tasks autonomously. The EVO-X5 Pro is positioned to become a dominant force in this new hardware category. While official pricing has not yet been confirmed, early-generation models utilizing the Ryzen AI Max+ 395—such as the EVO-X3 and EVO-X2—debuted between \$1,799 and \$2,999. Consequently, a fully configured EVO-X5 Pro with 192GB of memory is expected to target the professional tier, retailing at several thousand dollars.

Nonetheless, building a traditional desktop setup with equivalent VRAM requires multiple high-end discrete GPUs, such as the NVIDIA RTX 5090 (retailing for \$2,500 to \$3,400 each), which demands massive space, specialized cooling, and substantial electrical infrastructure. By contrast, the EVO-X5 Pro delivers a highly efficient 120W to 140W solution that is quiet, space-saving, and economically sensible for developers and mid-sized enterprises. Its release stands as a vital milestone, accelerating the democratization of local AI development worldwide.

References

• 【GMKtec EVO-X5 Pro Official Announcement】https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc

【#MiniPC #GMKtec #LocalLLM #Ryzen #UnifiedMemory #AIWorkstation #AgenticPC】