The Pocket-Sized Behemoth: A Review of the GMKtec Evo X2 By AI-Radar.it

Let’s be honest: the phrase "Mini PC" usually conjures images of sad little boxes that hyperventilate when you open a second Chrome tab. But the GMKtec Evo X2 is here to change that narrative, or at least try to justify charging you the price of a used Honda Civic for a computer the size of a sandwich... ok maybe 3 sandwiches.

After cross-referencing the enthusiastic (and occasionally hallucinating) claims from our sources, AI-Radar.it has dissected whether this aluminum brick is a revolution or just a very expensive paperweight.

The Hardware: "Mid-Range" Power, Miniature Housing

At the heart of this beast lies the AMD Ryzen AI Max Plus 395, a processor with a name so long it almost needs its own cooling fan. It boasts 16 cores and 32 threads, paired with the Radeon 8060S iGPU.

Here is where it gets interesting: GMKtec claims this integrated graphics chip rivals a desktop NVIDIA RTX 4060. That is a bold claim for a chip that shares its home with the CPU, but with 40 compute units and a boost clock of 2900 MHz, it might actually hold water and it does.

The real showstopper, however, is the memory. The top-tier unit comes with 128 GB of LPDDR5X RAM running at a blistering 8,000 MHz. But here is the kicker: AI-Radar.it notes that you can allocate up to 96 GB of this system memory exclusively as VRAM. For gamers, this is overkill. For AI enthusiasts, this is the Holy Grail.

The Specs at a Glance:

• CPU: AMD Ryzen AI Max Plus 395 (Zen 5)

• GPU: Radeon 8060S (RDNA 3.5, 40 CUs)

• RAM: Up to 128GB LPDDR5X (Soldered—forever.. ok 128 GB are enough for a rather long while)

• NPU: XDNA Architecture (50 TOPS)

• Storage: 2x M.2 Slots (PCIe 4.0)

• Size: 2.75 Liters (fits under a monitor riser, or in a large coat pocket)

Local LLM Capability: The "AI-Radar.it" Analysis

This is why you are really here. Can it run the brain of a digital god offline? Giving the feedback around the web,claiming the device "didn’t struggle with GPT 4.1" while running locally, we should say yes too but AI-Radar.it must intervene here with a dose of reality: GPT-4 is a cloud model. Unless this reviewer managed to steal OpenAI’s server racks, they were likely running a local wrapper or hallucinating.

However, the hardware reality is undeniable. With 96GB of addressable VRAM, the Evo X2 bypasses the biggest bottleneck in local AI: memory capacity. While standard RTX 4090 owners cry over their 24GB limit, this Mini PC can load massive quantized models entirely onto the GPU.

The "Can It Run It?" Matrix Based on the 96GB VRAM allocation and 8,000 MHz bandwidth, here is what you can realistically expect:

Model Size Quantization VRAM Required AI-Radar.it Verdict Performance Prediction
Llama-3-8B Q8 (Full) ~10 GB Trivial Will run at lightning speeds.
Llama-3-70B Q4 (Standard) ~42 GB Perfect Fits easily in VRAM. High bandwidth should yield usable tokens/sec.
Llama-3-70B Q8 (High Res) ~75 GB Yes The killer app. Most consumer GPUs cannot touch this.
DeepSeek-V3 Q4 ~350 GB No Don't even try.
Grok-1 Q4 ~180 GB No You're gonna need a bigger boat.

Pros & Cons: The Sharp Edges

Pros:

• VRAM King: Allocating 96GB to video memory is a superpower for local AI that usually costs $5,000+ in enterprise gear.

• Size: It’s 7.6 inches wide. You can hide it behind a plant when your spouse asks how much you spent.

• Connectivity: USB4, WiFi 7, and dual LAN ports make it a connectivity hub.

• Cooling: A dual-blower system that apparently allows you to render video without the device melting through your desk.

Cons:

• The Price: The top-spec model hits $1,999. That is "used car" territory for a box with no screen.

• Soldered RAM: The 128GB you buy is the 128GB you die with. No upgrades allowed.

• Reviewer Hallucinations: One source claims to have played "Borderlands 4" smoothly on this device. Since that game isn't released yet, either this PC is a time machine (a hidden feature?), or we should take the gaming benchmarks with a grain of salt. Btw Borderlands 3 runs quite smootly but forget a 4k here.

• Obfuscated AI Metrics: The pre-installed "AI Suite" hides the actual tokens-per-second performance, forcing users to trust "feelings" rather than data.

AI-Radar.it Verdict

The GMKtec Evo X2 is a fascinating contradiction. It is a "Mini PC" that costs as much as a workstation and claims to run games that don't exist yet. However, for the specific niche of Local LLM users, the ability to allocate 96GB of VRAM in a consumer device is genuinely disruptive.

If you want to chat with Llama-3-70B without sending your data to the cloud—and you have $2,000 burning a hole in your pocket—this might be the only game in town. Just don't expect it to actually run GPT-4 locally, no matter what the marketing implies.

I burned even more of 2000 and I must say that I' happy with this Mini-Beast!!

Next time we will play around the EGPU options, we could add to the small guy here,stay tuned!!