pickwise

Intel · Local AI machine

GMKtec EVO-T2 (64 GB)

Intel machine: Intel Core Ultra X7 358H (Panther Lake, Arc B390, 12 Xe3 cores), 64 GB memory, 36.5 GB for the GPU, 136.5 GB/s memory speed.

77.8Score

  • #33 of 41 ranked
  • #1 of 4 Intel machines

Why this score

The score adds up these parts. Each part has a weight. A part counts as the average of its sources, pulled toward 80 when there are few sources.

  1. AI power

    50% weightcounts as 73.1

    Average of 2 sources: 69.7

  2. Experts

    30% weightcounts as 83.9

    Average of 2 sources: 85.9

  3. Users

    20% weightcounts as 80.0

    No sources found yet. No data yet. This part counts as 80.

Thin: only 2 reviews and buyer ratings, so they are pulled toward the average.

80 is the field average. A part counts as 80 + (average − 80) × N ÷ (N + 1), where N is its number of sources. One source counts half of the gap from 80. No sources count as 80.

Price

€1799.99 at GMKtec EU shop (de.gmktec.com) · Maker's EU shop, ships to Austria

Price shown on the maker's EU page for all listed EVO-T2 options (X7 358H 64 GB / 1 TB and X9 388H variants); two options marked sold out on the page. Shipping to Austria not tested at checkout.

Prices checked 8 Oct 2026.

Specs

Chip
Intel Core Ultra X7 358H (Panther Lake, Arc B390, 12 Xe3 cores)
CPU
16 cores / 16 threads (4P + 8E + 4LP-E)
Memory
64 GB
GPU memory
36.5 GBWindows: Intel Shared GPU Memory Override, default 57% of RAM (Intel support article; the maximum depends on RAM and is not stated). 57% x 64 GB = 36.5 GB. Press reports (mezha.ua) say Intel driver 32.0.101.8804 allows up to 93% (59.5 GB) on Core Ultra Series 3, described for Arc Pro B370/B390; Intel itself does not document it, so it is not used. No Linux limit found.
Memory speed
136.5 GB/s
FP16 TFLOPS
—
NPU
50 TOPS
OS
Windows, Linux
Software
Vulkan
Ethernet
10 Gbit/s
Thunderbolt / USB4
USB4
QSFP port
No
Idle power
13 W
Load power
72 W
Noise
38 dB(A)
Volume
1.7 L
Weight
1 kg
SSD
1 TB
Release
2026-03

Measured

Qwen3.6-35B-A3B, not stated (GMKtec 'Balanced' preset)
30 tok/s writing, 500 tok/s prompt · GMKtec bundled local-AI app, GPU only (backend not named)

Notes

Arc B390 plus fast soldered LPDDR5X (about 137 GB/s) runs a 35B mixture-of-experts model at roughly 30 tokens/s on the GPU alone, and it is quiet with 10 GbE. The 64 GB is soldered, Intel's default GPU share is about 36.5 GB, and there are no 70B-class or gpt-oss-120b numbers.

Memory speed: LPDDR5X-8533 on the maker EU page, in Tom's spec table and in TechRadar's table, but Tom's text says the US configs are LPDDR5X-8000 (128 GB/s); bw uses 8533. Weight: 950 g (maker) vs 937 g (Tom's). TechRadar's spec table lists 103 x 98 x 32 mm and 1,273 g with PSU for the T2, apparently copied from the T1; the maker and Tom's dimensions are used. TechRadar says its review unit shipped with a single 16 GB module, while Tom's tested 64 GB. Tom's pp '500+' is a lower bound. GPU memory uses the 57% share Intel documents; press figures for higher shares are in the GPU memory note.

More reading:starryhope.com

See it in the Local AI Index