GMKtec EVO-X5 Pro Launches September 28 With 192GB Unified Memory

GMKtec says its EVO-X5 Pro desktop will officially launch on September 28, 2026. The company published the date on September 15 after unveiling the system at IFA earlier in the month, turning what had been an announced product into a launch with a specific near-term date.

The system is built around AMD’s Ryzen AI Max+ PRO 495 and is positioned primarily for local AI and professional workloads. GMKtec says configurations can include up to 192GB of unified LPDDR5X memory, an unusually large memory pool for a compact desktop and the feature most directly relevant to running large local models.

AMD’s current specification page for the Ryzen AI Max+ PRO 495 lists 16 Zen 5 CPU cores and 32 threads, Radeon 8065S integrated graphics with 40 graphics cores, an XDNA 2 NPU rated at up to 55 TOPS, and support for as much as 192GB of 256-bit LPDDR5X-8533 memory. AMD lists the processor for both laptop and desktop form factors.

That matters because the EVO-X5 Pro’s large-memory positioning is not based only on a system-vendor marketing claim. The 192GB platform ceiling is part of AMD’s own processor specification. The exact memory configuration, cooling, firmware and sustained performance of the finished GMKtec system remain product-specific questions rather than properties guaranteed by the processor specification alone.

Large local language models are often constrained by memory capacity before raw compute becomes the only problem. A larger unified pool can make room for quantized models that do not fit on conventional consumer GPUs with much smaller dedicated VRAM capacities. AMD itself highlights the Ryzen AI Max+ PRO 495 platform’s ability to allocate up to 160GB as graphics memory for very large local models.

That does not make unified memory equivalent to high-end discrete-GPU HBM or GDDR memory, nor does it prove comparable inference throughput. Capacity, bandwidth and compute are separate constraints. The useful distinction is that the EVO-X5 Pro is designed to make a very large addressable memory pool available to CPU and integrated-GPU workloads in one desktop system.

GMKtec’s pre-launch material names the date but does not yet provide enough evidence to treat every configuration, price or region as settled. The launch should clarify which memory and storage configurations are actually orderable, their regional pricing, and when buyers can expect systems to ship.

For local-AI users, independent testing will matter after hardware becomes available. The most useful results will separate model size from real inference speed, memory use, sustained power and thermals, software compatibility and output quality instead of treating the ability to load a large quantized model as a complete performance result.