GMKTEC EVO-X2 (STRIX HALO)

BUY NOW AT GMKTEC →

product: GMKtec EVO-X2 (Strix Halo)
category: Mini PCs · also under Workstations
brand: GMKtec
memory: 128GB unified · 256 GB/s [2]
notes: Maker store price, read from Wayback snapshots of the first-party store page (cheapest 128GB variant).
price:
  now: $3,499.99 (GMKtec, official list, 2026-10-15) [1] BUY NOW AT GMKTEC →
  since_first: UP 6.1% SINCE MAY 29, 2026
  first_record: 2026-05-29 (price in the paid export)
  change:
    1d: —
    7d: —
    14d: —
    30d: =0.0%
    90d: =0.0%
    180d: new
  price_hikes: 1 since May 2026

UP 6.1% SINCE MAY 29, 2026 — now $3,499.99. BUY NOW AT GMKTEC →
1 PRICE HIKE SINCE MAY 2026
LATEST STEP UP: +$200 (+6.1%), FIRST SEEN JUL 21, 2026

PRICE TREND (shape since first record)

now $3,499.99 │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │                   █████████████████████████████████████████
              │███████████████░░░░█████████████████████████████████████████
              └────────────────────────────────────────────────────────────
               2026-05-29                                        2026-11-13
               █ recorded price  ░ gap between records · shape only; past prices are in the paid export

PRICE HISTORY (last 7 days)

no price recorded in the last 7 days; the current price above is the latest record

9 older records back to May 29, 2026 (every price, date and source) are in the paid export (YAML / CSV).

PAIRS WELL WITH (add-ons, not price-tracked · all peripherals)
Samsung 990 PRO 4TB NVMe SSD (MZ-V9P4T0B/AM):
  why: Model files are big: a 70B Q8 file alone is ~75 GB. Framework Desktop DIY ships without storage.
  buy: BUY ON AMAZON →
CalDigit TS4 Thunderbolt 4 / USB4 dock:
  why: Adds ports to USB4 Strix Halo boxes and Macs; 18 ports, 98W charging.
  buy: BUY ON AMAZON →
Plugable USB-C 2.5Gb Ethernet adapter (USBC-E2500):
  why: Cheap 2.5GbE for machines with 1GbE or Wi-Fi only.
  buy: BUY ON AMAZON →
APC Back-UPS Pro 1500VA (BR1500MS2):
  why: Keeps a long run or a model download alive through a short outage.
  buy: BUY ON AMAZON →

CAN RUN — LOCAL MODELS THAT FIT

memory: 128GB unified · usable for models ≈ 96 GB (rule)
bandwidth: 256 GB/s [2]
largest_fit: Nemotron 3.5 Super 120B Q4_0
fits: 29 of 33 model files (excluding the Llama 2 7B reference)
note: fit = weights + KV cache at 8,192 tokens + 10% runtime overhead
Nemotron 3.5 Super 120B-A12B (MoE)Q8_0128.5 GB141.4 GB [3]nono published run
Nemotron 3.5 Super 120B-A12B (MoE)Q4_072.1 GB79.4 GB [3]YESno published run
gpt-oss 120B (MoE)MXFP463.4 GB70.0 GB [5]YES47.5 tok/s tg32 [4]
Llama 4 Scout 109B-A17B (MoE)Q8_0114.5 GB127.6 GB [6]TIGHTno published run
Llama 4 Scout 109B-A17B (MoE)Q4_K_M65.4 GB73.5 GB [6]YESno published run
Nemotron 3.5 Bolt 74B-A9B (MoE)BF16148.2 GB163.1 GB [7]nono published run
Nemotron 3.5 Bolt 74B-A9B (MoE)Q8_078.9 GB86.8 GB [7]YESno published run
Nemotron 3.5 Bolt 74B-A9B (MoE)Q4_K_XL50.2 GB55.3 GB [7]YESno published run
Llama 3.3 70Bf16141.1 GB157.9 GB [8]nono published run
Llama 3.3 70BQ8_075.0 GB85.2 GB [8]YESno published run
Llama 3.3 70BQ4_K_M42.5 GB49.5 GB [8]YESno published run
OLMo 3.1 32Bbf1664.5 GB72.3 GB [9]YESno published run
OLMo 3.1 32BQ8_034.3 GB39.0 GB [9]YESno published run
OLMo 3.1 32BQ4_K_M19.5 GB22.8 GB [9]YESno published run
Nemotron 3.5 Lightning 30B-A3B (MoE)BF1663.2 GB69.5 GB [10]YESno published run
Nemotron 3.5 Lightning 30B-A3B (MoE)Q8_033.6 GB37.0 GB [10]YESno published run
Nemotron 3.5 Lightning 30B-A3B (MoE)Q4_018.9 GB20.8 GB [10]YESno published run
Gemma 4 31BBF1661.4 GB69.1 GB [11]YESno published run
Gemma 4 31BQ8_032.6 GB37.4 GB [11]YESno published run
Gemma 4 31BQ4_K_M18.3 GB21.7 GB [11]YESno published run
Granite 4.2 30Bbf1658.6 GB66.6 GB [12]YESno published run
Granite 4.2 30BQ8_031.1 GB36.4 GB [12]YESno published run
Granite 4.2 30BQ4_K_M17.7 GB21.6 GB [12]YESno published run
Gemma 4 26B-A4B (MoE)BF1650.5 GB55.9 GB [13]YESno published run
Gemma 4 26B-A4B (MoE)Q8_026.9 GB29.9 GB [13]YESno published run
Gemma 4 26B-A4B (MoE)Q4_K_M16.9 GB19.0 GB [13]YESno published run
gpt-oss 20BMXFP412.1 GB13.5 GB [14]YESno published run
Phi-4 14BF1629.3 GB33.9 GB [15]YESno published run
Phi-4 14BQ8_015.6 GB18.8 GB [15]YESno published run
Phi-4 14BQ4_K_M8.9 GB11.5 GB [15]YESno published run
Gemma 4 12BBF1623.8 GB26.7 GB [16]YESno published run
Gemma 4 12BQ8_012.7 GB14.4 GB [16]YESno published run
Gemma 4 12BQ4_K_M7.1 GB8.3 GB [16]YESno published run
Llama 2 7B (llama.cpp reference benchmark model)Q8_07.2 GB12.2 GB [17]YESno published run
Llama 2 7B (llama.cpp reference benchmark model)Q4_03.8 GB8.5 GB [17]YES51.8 tok/s tg128 [18]

WHERE TO BUY

official: BUY NOW → GMKtec
retailers: not tracked yet (partner programs pending)
TRACK PRICES — WEEKLY EMAIL COMING SOON
Get one email a week with the price changes for the products you pick. Not live yet: this form sends nothing and stores nothing.




SOURCES

Current sources only. Every older price point and its source is in the paid export.

#whatsourcefetched
[1]GMKtec EVO-X2 $3,499.99 on 2026-10-15 — Wayback snapshot 20261015 of https://www.gmktec.com/products/amd-ry...web.archive.org/web/20261015170511/www.gmktec.com/products/amd-ryze...2026-11-13
[2]Ryzen AI Max+ 395 platform figure (Framework blog); maker page not checkedframe.work/blog/introducing-the-framework-desktop2026-11-13
[3]Nemotron 3.5 Super 120B-A12B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/ggml-org/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF2026-11-13
[4]GMKtec EVO-X2 (Ryzen AI Max+ 395, 128GB), ROCm, posted in llama.cpp #16578github.com/ggml-org/llama.cpp/discussions/165782026-11-13
[5]gpt-oss 120B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/ggml-org/gpt-oss-120b-GGUF2026-11-13
[6]Llama 4 Scout 109B-A17B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF2026-11-13
[7]Nemotron 3.5 Bolt 74B-A9B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF2026-11-13
[8]Llama 3.3 70B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/bartowski/Llama-3.3-70B-Instruct-GGUF2026-11-13
[9]OLMo 3.1 32B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/bartowski/allenai_Olmo-3.1-32B-Instruct-GGUF2026-11-13
[10]Nemotron 3.5 Lightning 30B-A3B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF2026-11-13
[11]Gemma 4 31B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/gemma-4-31B-it-GGUF2026-11-13
[12]Granite 4.2 30B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/ibm-granite/granite-4.2-30b-GGUF2026-11-13
[13]Gemma 4 26B-A4B (MoE): GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF2026-11-13
[14]gpt-oss 20B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/ggml-org/gpt-oss-20b-GGUF2026-11-13
[15]Phi-4 14B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/phi-4-GGUF2026-11-13
[16]Gemma 4 12B: GGUF sizes + config.json (KV cache at 8k)huggingface.co/unsloth/gemma-4-12b-it-GGUF2026-11-13
[17]Llama 2 7B (llama.cpp reference benchmark model): GGUF sizes + config.json (KV cache at 8k)huggingface.co/TheBloke/Llama-2-7B-GGUF2026-11-13
[18]Vulkan scoreboard row 'AMD Ryzen AI Max+ 300 Series' (chip-level result, not this exact box)github.com/ggml-org/llama.cpp/discussions/108792026-11-13