Skip to main content
Mini PC Lab logo
Mini PC Lab Tested. Benchmarked. Reviewed.
comparisons

GMKtec EVO-X2 vs GEEKOM A9 Max: Strix Halo 128GB vs Strix Point — $1,300 Upgrade Question [2026]

By Mini PC Lab Team · January 27, 2026 · Updated February 1, 2026

This article contains affiliate links. If you purchase through our links, we may earn a commission at no extra cost to you. We only recommend products we’ve thoroughly researched.

GMKtec EVO-X2 vs GEEKOM A9 Max: Strix Halo 128GB vs Strix Point — $1,300 Upgrade Question [2026]

GMKtec EVO-X2 AI

The Short Answer

The GMKtec EVO-X2 at $2,999 is the only mini PC that can run 70B+ parameter LLMs comfortably — its 128GB LPDDR5X RAM with 96GB VRAM allocation makes this possible. The GEEKOM A9 Max at $1,689 maxes out at ~13B models with 32GB RAM.

If running large language models locally is your primary use case, the EVO-X2 is worth the $1,300 premium. If you only run 7B-13B models or want a balanced AI + homelab + daily driver, the A9 Max saves you $1,300 and gives you a 3-year warranty.

Side-by-Side Specs

SpecGMKtec EVO-X2 AIGEEKOM A9 MaxWinner
CPURyzen AI Max+ 395 (16C/32T, 5.1GHz, Strix Halo)Ryzen AI 9 HX 370 (12C/24T, 5.1GHz, Strix Point)🏆 EVO-X2 (Strix Halo)
GPURadeon 8060S (40 CUs, 2,560 shaders)Radeon 890M (16 CUs, 1,024 shaders)🏆 EVO-X2 (2.5x more CUs)
RAM128GB LPDDR5X 8000MT/s (soldered, 8-channel)32GB DDR5 SO-DIMM (upgradeable to 128GB)Contextual
Storage2TB PCIe NVMe SSD1TB PCIe NVMe SSD (dual M.2)🏆 EVO-X2 (more included)
Networking2.5GbE (Realtek)Dual 2.5GbE (Intel)🏆 A9 Max (Intel NICs)
WiFiWiFi 7WiFi 7Tie
AI TOPS126 (50+ NPU + GPU)80 (50 NPU + 30 GPU)🏆 EVO-X2
VRAM AllocationUp to 96GB to GPU via BIOSShared system memory🏆 EVO-X2
Warranty1 year3 years🏆 A9 Max
Reviews74 (4.1★)106 (4.4★)🏆 A9 Max (more feedback)
Price~$2,999~$1,689🏆 A9 Max ($1,310 savings)

Power Consumption

MetricGMKtec EVO-X2 AIGEEKOM A9 Max
Idle (W)~12W~9W
Load (W)~120W~80W
Annual Cost (24/7 idle)~$12.61/year~$9.46/year

Annual cost calculated at $0.12/kWh, running 24/7 at idle. Sources: ServeTheHome (Strix Halo platform) and NotebookCheck (Strix Point platform).

The EVO-X2 draws more power because the Strix Halo platform is fundamentally more powerful — 128GB LPDDR5X memory and 40-CU GPU contribute to higher baseline consumption. For always-on use, that’s an extra ~$3/year in electricity.

Detailed Breakdown

CPU Performance: Strix Halo vs Strix Point

The Ryzen AI Max+ 395 (Strix Halo) has 16 Zen 5 cores and 32 threads. The Ryzen AI 9 HX 370 (Strix Point) has 12 Zen 4 cores and 24 threads.

For multi-threaded workloads:

  • The 395 wins by roughly 25-30% in Cinebench R23
  • Video encoding, 3D rendering, and batch processing benefit significantly
  • Running 10+ VMs simultaneously favors the 395

For single-threaded workloads:

  • Both chips are competitive — same 5.1 GHz boost clock
  • Desktop tasks feel identical on both systems
  • Neither chip bottlenecks everyday use

GPU: 40 CUs vs 16 CUs

This is the biggest differentiator after RAM.

Radeon 8060S (EVO-X2):

  • 40 RDNA 3.5 compute units (2,560 shaders)
  • Performance between RTX 4060 and RTX 4070 laptop GPUs
  • Handles 1080p gaming at ultra settings (60+ fps in AAA titles)
  • Generates SDXL images in ~8-12 seconds
  • Can allocate up to 96GB VRAM for large LLM inference

Radeon 890M (A9 Max):

  • 16 RDNA 3.5 compute units (1,024 shaders)
  • Performance comparable to entry-level discrete GPU
  • Handles 1080p gaming at medium-high settings (40-60 fps)
  • Generates SDXL images in ~20-30 seconds
  • Shares system RAM (up to 32GB without upgrade)

For gaming: The 8060S is 2-2.5x faster. If you care about 1080p ultra gaming, the EVO-X2 is in a different league.

For Stable Diffusion: The 8060S generates images 2-3x faster. For serious AI art work, this matters.

For LLM inference: The 8060S accelerates token generation, but RAM capacity is the limiting factor for model size.

RAM: The Deciding Factor

EVO-X2 (128GB LPDDR5X soldered):

  • 8-channel memory architecture (~256 GB/s bandwidth)
  • Cannot be upgraded — what you buy is what you keep
  • Enables running 70B+ parameter LLMs locally (Qwen 72B, DeepSeek 70B, Llama 3.1 70B)
  • User benchmarks confirm: Qwen3 235B at 8-10 tokens/sec, gpt-oss-120b at 36-40 tokens/sec

A9 Max (32GB DDR5 SO-DIMM upgradeable):

  • Dual-channel memory architecture (~100 GB/s bandwidth)
  • Can upgrade to 64GB, 96GB, or 128GB later (2x DIMM slots)
  • Handles 7B-13B models comfortably with 32GB
  • Can run 34B models at 32GB with quantization
  • Requires RAM upgrade (~$200-400) for 70B models

LLM RAM requirements (Q4 quantization):

Model SizeRAM Required
7B~4-6GB
13B~8-10GB
34B~20-24GB
70B~42-48GB
120B+~70-80GB

Verdict: The EVO-X2’s 128GB is essential for 70B+ models. The A9 Max can get there with a RAM upgrade, but you’re still limited by dual-channel bandwidth.

Networking: Dual Intel 2.5GbE vs Single Realtek

A9 Max:

  • Dual Intel 2.5GbE NICs (i226-V)
  • Perfect for OPNsense/pfSense firewall (WAN + LAN)
  • Well-supported in Linux and Proxmox
  • Intel drivers are mature and reliable

EVO-X2:

  • Single 2.5GbE (Realtek controller)
  • Adequate for general use
  • Realtek drivers work on Linux but occasionally require manual installation
  • Not ideal for firewall/router use

For homelab: The A9 Max is the better choice if you need multi-NIC setups.

Warranty: 3 Years vs 1 Year

A9 Max: 3-year limited warranty — the best in the mini PC industry. GEEKOM has a track record of honoring warranties and providing BIOS updates.

EVO-X2: 1-year limited warranty — standard for the industry but not exceptional.

If you’re risk-averse or plan to run 24/7 for years, the 3-year warranty is meaningful. At $2,999, the EVO-X2’s 1-year warranty feels thin.

Real-World User Benchmarks

From EVO-X2 customer reviews (Edward Lee, verified purchaser):

Qwen3 235B (235 billion parameters):

  • 8-10 tokens/sec with ROCm-enabled llama.cpp
  • LM Studio settings: 96GB VRAM allocation, 8000MT/s memory

gpt-oss-120b (120 billion parameters):

  • 36-40 tokens/sec
  • Same settings as above

These are real-world numbers from an actual user, not manufacturer claims. They confirm the EVO-X2 handles models that would choke any other mini PC on the market.

Price and Value

At $2,999, the EVO-X2 costs $1,310 more than the A9 Max at $1,689. What does that premium buy you?

EVO-X2 advantages:

  • +4 CPU cores (16C vs 12C)
  • +24 GPU CUs (40 vs 16 — 2.5x more)
  • +96GB RAM (128GB vs 32GB)
  • 8-channel memory bandwidth (~256 GB/s vs ~100 GB/s)
  • +46 TOPS AI (126 vs 80)
  • Up to 96GB VRAM allocation

A9 Max advantages:

  • Upgradeable RAM (can reach 128GB later with DDR5)
  • Dual Intel 2.5GbE NICs
  • 3-year warranty
  • $1,310 savings

Value analysis:

  • If you need 128GB RAM for 70B+ LLMs — the EVO-X2 is worth it (this is the only mini PC that does this comfortably)
  • If you only run 7B-13B models — the A9 Max is overkill for your needs
  • If you want to upgrade RAM later — the A9 Max allows this (~$200-400 for 64GB kit)
  • If warranty matters — the A9 Max’s 3 years vs 1 year is significant

When the EVO-X2 Is Worth $1,300 More

Buy the EVO-X2 if:

  • You need to run 70B+ parameter LLMs locally (this is the killer use case)
  • You do serious GPU compute work (Stable Diffusion at scale, video editing, 3D rendering)
  • You want the most powerful mini PC on the market (bragging rights)
  • You have the budget and don’t need to justify the expense

The EVO-X2 is NOT worth it if:

  • You only run 7B-13B models (32GB is plenty)
  • You’re on a budget (the A9 Max or cheaper options work fine)
  • You need dual 2.5GbE for homelab firewall use
  • You want a 3-year warranty for peace of mind

Final Verdict: Which Should You Buy?

Buy the GMKtec EVO-X2 AI if:

  • Running 70B+ LLMs locally is your primary use case (this is the only mini PC that does this well)
  • You want desktop-class GPU performance (40 RDNA 3.5 CUs)
  • You do serious AI development or GPU compute work
  • Budget is not a constraint

Buy the GEEKOM A9 Max if:

  • You run 7B-34B models (32GB is sufficient, or upgrade to 64GB later)
  • You want a balanced AI + homelab + daily driver
  • You value dual Intel 2.5GbE for networking
  • You want a 3-year warranty
  • You’d rather save $1,310 and spend it on RAM, storage, or peripherals

Our pick: For most users, the GEEKOM A9 Max at $1,689 is the smarter choice. It handles 7B-34B models comfortably, has upgradeable RAM, dual Intel 2.5GbE, and a 3-year warranty. The $1,310 savings can buy a 64GB RAM upgrade, a 2TB NVMe, and still leave money for peripherals.

The EVO-X2 only makes sense if you specifically need 128GB RAM for 70B+ LLMs — and if you do, you already know which one to buy.