Local Boxes / directory

Buy the box.
Keep the weights.

A slow directory of computers that run real models on your desk. We list what it is, who it is for, what it actually runs, and the catch.

Listings: 10 Hubs: 1 Comparisons: 1 Next ship: Used RTX 5090 SFF workstation (as a class, not one SKU)
Hub Jetson and boxes under $500 The only listing that belongs here today is the Orin Nano Super. Compare DGX Spark vs ASUS GX10 Same GB10 class. Two invoices. Shop them as one product.

HP Z2 Mini G1a

For people who want Strix Halo in a certified, mountable mini workstation with HP support options.

Street
$2,300–$5,000+
Memory
up to 128GB LPDDR5X
Runs
24B–70B class
Catch
Soldered; 2.5GbE; enterprise pricing

The enterprise Strix Halo mini. Same bandwidth ceiling as the sealed cubes. Buy 128GB.

Jetson AGX Thor

For robotics and physical-AI teams who need Blackwell + 128GB under JetPack.

Street
$3,499–$5,499
Memory
128GB LPDDR5X
Runs
20B–120B MoE class
Catch
Price jumped; robotics-first

High-end Jetson for physical AI. Same memory class as Spark, Arm + sensor I/O.

Beelink GTR9 Pro

For people who want Strix Halo with dual 10GbE in a sealed Mac Studio–shaped cube.

Street
$3,600–$4,350
Memory
128GB LPDDR5X
Runs
24B–70B class
Catch
Soldered; price rose

The dual-10GbE Strix Halo cube. Same bandwidth ceiling as Framework and EVO-X2. Confirm NIC revision.

Minisforum MS-S1 MAX

For people who want Strix Halo with dual 10GbE and a limited PCIe slot in one chassis.

Street
$2,599–$3,799
Memory
64–128GB LPDDR5X
Runs
24B–70B class
Catch
Soldered; PCIe is x4

The networked Strix Halo mini. Buy 128GB. Same bandwidth ceiling as Framework and EVO-X2.

Mac mini M6 / M5 Pro

For macOS users who want a silent 14B–27B (or 64GB M5 Pro) box without leaving Apple’s stack.

Street
$899–$1,699+
Memory
16–64GB unified
Runs
3B–70B class
Catch
Soldered; price jumped

Current shipping Mac mini. Buy configured RAM now. Not CUDA, not 128GB.

Jetson Orin Nano Super

For students, robots, and anyone who needs CUDA at the edge for $249.

Street
$249 kit
Memory
8GB LPDDR5
Runs
1B–8B class
Catch
8GB is the ceiling

The cheap door into JetPack and physical AI. Not a 70B desk box.

NVIDIA DGX Spark

For CUDA people who want 128GB of unified memory in a paperback-sized brick.

Street
~$4,699
Memory
128GB LPDDR5X
Runs
20B–120B MoE locally
Catch
Bandwidth is the limiter

Buy this if you already live in CUDA and need capacity, not the fastest tokens per dollar.

Framework Desktop

For people who want a repairable Strix Halo box they can live with for five years.

Street
$1,269–$3,449
Memory
up to 128GB unified
Runs
24B–70B class
Catch
You assemble it

The grown-up midrange pick. Not CUDA-native.

GMKtec EVO-X2

For people who want Strix Halo already assembled, in a 193 mm mini, and will pay for 128GB.

Street
$3,500–$3,650
Memory
128GB LPDDR5X
Runs
24B–70B class
Catch
Sealed; $2,200 is 64GB

The sealed midrange mini. Same chip as Framework. Buy 128GB or skip it.

ASUS Ascent GX10

For buyers who want the DGX Spark recipe without the Founders Edition.

Street
~$3,100–$4,150
Memory
128GB unified
Runs
GB10 / CUDA stack
Catch
Partner firmware lottery

Same class as Spark, often cheaper. Read the SKU.