GMKtec EVO-X2 Review: Is This Mini PC Actually a Local AI Workstation?

What Is the GMKtec EVO-X2?

The GMKtec EVO-X2 is a compact desktop computer built around AMD's Ryzen AI Max+ 395 processor, code-named “Strix Halo,” designed to bring workstation-class CPU, GPU and memory bandwidth into a footprint smaller than a shoebox. Instead of a discrete graphics card, it uses AMD's integrated Radeon 8060S GPU paired with unified LPDDR5X memory — meaning the same fast memory pool serves both the CPU and GPU, a design borrowed from high-end laptops and now applied to a desktop form factor.

It ships in configurations from 64GB up to 128GB of memory and up to 2TB of PCIe 4.0 SSD storage, with a body measuring roughly 193 x 186 x 77mm — small enough to sit beside a monitor rather than under a desk.

Full Specifications

The Feature That Actually Matters: Unified Memory

The single most important spec on the EVO-X2 is not the CPU clock speed — it is how much of its memory the integrated GPU can address for AI workloads. Because Strix Halo uses unified memory, the 128GB configuration can allocate up to 96GB of that pool directly to the GPU, which is an extraordinary amount of GPU-addressable memory for a machine in this price bracket. For context, a single discrete workstation GPU with 48GB of memory, such as an NVIDIA RTX 6000-class card, typically costs more on its own than an entire EVO-X2 unit.

That matters specifically for running large language models locally: model size that can be loaded is generally bound by available GPU memory, so a 96GB-addressable unified pool can run substantially larger open-weight models on-device than a typical consumer GPU with 8–24GB of dedicated VRAM, without the latency or data-residency concerns of calling a cloud API.

Benchmark Numbers Worth Knowing

Independent lab testing (Tom's Hardware, 64GB configuration) recorded 90–110 fps in Star Wars Jedi: Fallen Order at 1080p Epic settings, and a stable 57 fps average across 15 sequential runs of the demanding Metro Exodus benchmark, with less than a 1 fps variation between runs — indicating consistent thermal performance under sustained load. CPU temperature during that stress test averaged 61°C.

In productivity workloads — Handbrake video encoding and Geekbench CPU tests — the EVO-X2 performed almost identically to the Framework Desktop, which uses the same Ryzen AI Max+ 395 chip, confirming that raw CPU performance is consistent across vendors building on this platform. The one weak spot in third-party testing was file transfer speed on a 25GB test file, which underperformed expectations for a fully-equipped system.

Trade-offs to Know Before You Buy

• Memory and CPU are soldered. Only the two M.2 storage slots are user-upgradeable — decide on 64GB, 96GB or 128GB at purchase, because you cannot add memory later.
• External power brick. The 230W adapter is bulkier than the machine itself, which slightly undercuts the space-saving pitch versus a true all-in-one design.
• Fan noise under sustained load. Independent reviewers noted audible fan noise during CPU-heavy benchmarks, though noise was better controlled during gaming since the CPU is not fully stressed.
• Firmware update cadence. As with any new Strix Halo platform, check the current BIOS/firmware revision before deploying it for anything security-sensitive, since early units have needed vendor firmware updates for platform security patches.

Who Should Actually Buy the EVO-X2

The EVO-X2 makes the most sense for two groups: developers and technical teams who want to run large open-weight AI models locally for privacy or cost reasons, and space-constrained creators who want strong CPU performance and credible 1080p gaming without a mid-tower case. If your workload genuinely needs the 128GB configuration's 96GB of GPU-addressable memory, it undercuts the price of comparable workstation hardware by a wide margin.

It makes less sense as a pure gaming PC — a mid-tower desktop with a discrete GPU will deliver better performance per dollar for gaming alone — and it is not the right choice if you need to upgrade memory down the line, since the memory is soldered at purchase.

“GMKtec's Evo X2 delivers impressive horsepower and credible 1080p muscle into a remarkably compact chassis. Its value hinges on your workload.”Tom's Hardware, GMKtec Evo X2 AI Mini PC review

Why This Matters for Singapore SMEs Exploring Local AI

For Singapore SMEs weighing whether to run AI workloads locally versus through a managed cloud service, unified-memory mini PCs like the EVO-X2 lower the cost of entry for experimentation and data-sensitive use cases considerably. That said, most SMEs are better served by a managed, supervised AI service for day-to-day operations — accounting, marketing, HR and admin — where reliability, human review and support matter more than owning the underlying hardware. Local hardware like the EVO-X2 is better suited to technical teams prototyping or fine-tuning models, not to running a small business's daily operations unsupervised.