Unveiled at the IFA trade fair in Berlin running from September 4 to 8, 2026, hardware manufacturer GMKtec’s EVO-X5 Pro is a high-end mini-PC engineered for professional developers running local artificial intelligence workflows. Powered by an AMD Ryzen AI Max+ 495 Pro processor featuring 16 Zen-5 cores and a 55 TOPS Neural Processing Unit, the system supports up to 192 GB of LPDDR5X-8533 memory, enabling the local execution of large language models.
Engineering High-Performance AI Compute in a Compact Chassis
The core of the EVO-X5 Pro relies on the AMD Ryzen AI Max+ 495 Pro silicon. This architecture packs 16 Zen-5 cores capable of boosting up to 5.2 GHz, paired with an integrated NPU delivering 55 TOPS of inference performance. The motherboard supports up to 192 GB of LPDDR5X-8533 RAM. Crucially, up to 160 GB of this capacity can be dynamically allocated as video RAM (VRAM), giving the compact form factor a memory pool that rivals workstation-class hardware. The subsystem sustains a memory bandwidth of 273 GB/s.
Thermal management for a small-form-factor machine running heavy workloads requires aggressive engineering. GMKtec deploys a vapor chamber cooling solution backed by a triple-fan array to pull heat away from the dense silicon. Users can manually tune the system’s thermal design power (TDP) profiles using dedicated function keys located directly on the chassis.
Running 300-Billion Parameter Models Locally
According to the manufacturer, the hardware can host LLMs containing up to 300 billion parameters if they use 4-bit quantization. During live demonstrations running the Qwen 3.8 model, the system processed between 18 and 20 tokens per second.
When shifting to denser 300B configurations, hardware constraints become apparent. The 273 GB/s memory bandwidth drops performance on dense models to roughly 1.5 tokens per second. However, Mixture-of-Experts (MoE) architectures are expected to yield significantly higher throughput due to their sparse activation patterns, making them ideal candidates for this localized hardware.
Enterprise Connectivity and Physical Interfaces
Deploying workstation power into enterprise environments demands robust physical security and network input-output capabilities. The front panel of the EVO-X5 Pro hosts two USB4 ports, while the rear features two USB4 V2 connections alongside dual 10GbE LAN ports for high-speed network integration. For remote administration and asset tracking, the system integrates enterprise management features including DASH and a Trusted Platform Module (TPM), and is built for continuous 24/7 operation.
Market Positioning and Pricing Context
GMKtec plans to launch the EVO-X5 Pro in black and silver chassis options starting in October 2026. While exact retail pricing has not been finalized at the time of the IFA showcase, historical pricing offers a clear baseline. The preceding EVO-X2 model equipped with 128 GB of RAM debuted at over 3,200 euros.
In the broader hardware landscape, GMKtec is carving out a niche beneath ultra-premium workstations. Apple’s Mac Studio M5 Ultra features 512 GB of unified memory with a massive 1.2 TB/s bandwidth, but commands a price tag exceeding 10,000 euros. Meanwhile, developer systems like the DGX Spark offer 128 GB of RAM and an identical 273 GB/s memory bandwidth at a valuation of roughly 4,000 US dollars. The EVO-X5 Pro targets developers who need near-workstation memory capacities without entering five-figure hardware budgets.
