NVIDIA DGX Spark AI Desktop 2 Pack
- 2 DGX Spark units with 1 connecting cable
- GB10 Grace Blackwell Superchip per unit
- Up to 1 PFLOPS of FP4 AI performance
- 128 GB unified memory per unit
- 4 TB NVMe M.2 SSD with self-encryption
- Models up to 200 billion parameters at FP4
- ConnectX-7 Smart NIC networking
- 12 x 8 x 6 inches, 7 pounds
Updated 25 days ago · Based on the Amazon listing and published specifications, not hands-on testing.
A pair of compact AI development computers linked by a cable for local model work. Built for developers on NVIDIA's software stack, not for gaming or office use.
As an Amazon Associate I earn from qualifying purchases. ProductDome earns a commission when you buy through links on this page.
ASIN: B0GWK5MPJ5Category: Mini Desktop Computers
Overview
Not a gaming desktop or a general office PC, the DGX Spark is a small AI workstation built on NVIDIA's Grace Blackwell superchip running DGX OS. This bundle pairs two units with a connecting cable so they can work together over the ConnectX network link.
Running and fine tuning large language models locally, without renting cloud hardware, is why developers buy this kit. It runs an Arm based Linux system with NVIDIA's own software, so Windows apps and games are off the table.
Who it's for
Model prototyping, fine tuning and inference on private data are the jobs here. People without Linux and AI framework experience will find little use for it.
Pros and cons
- Large unified memory for big models
- Two units linked for larger workloads
- Encrypted NVMe storage
- Full NVIDIA AI software stack
- Runs DGX OS, not Windows
- Arm processor limits general desktop software
- Only one HDMI output per listing
- Power plug type listed as none
Performance notes
Model size is governed by the unified memory, which the CPU and GPU share, so large models load without splitting across separate graphics memory. The headline compute figure reflects low precision inference, and higher precision work runs slower. Linking the two units through the network card lets a workload span both.
Specifications
| Brand | NVIDIA |
|---|---|
| Dimensions & fit | |
| Item dimensions | 12 x 8 x 6 inches |
| Capacity & performance | |
| Memory Storage Capacity | 4 TB |
| Processor Speed | 3.8 GHz |
| Compatibility & installation | |
| RAM Memory Installed | 128 GB |
| Connectivity Technology | Ethernet |
| Included components | |
| Unit Count | 2 Count |
| Warranty & support | |
| Warranty Description | 1 Year Limited |
| Other details | |
| Processor Series | GB10 Grace Blackwell Superchip |
| Operating System | NVIDIA DGX OS |
| Hard Disk Description | SSD |
| Total Usb Ports | 4 |
| Total Number of HDMI Ports | 1 |
| Personal Computer Design Type | Mini PC |
| CPU Model Number | ARM Cortex-X925 + ARM Cortex-A725 Processor |
| RAM Memory Technology | DDR5 |
| BuiltIn Media | 1 Connecting Cable, 2 DGX Spark Units, 2 Power Adapters |
Frequently asked questions
Can the DGX Spark run Windows?
The listing gives NVIDIA DGX OS as the operating system on an Arm processor, so plan on that system rather than Windows.
How large a model can DGX Spark handle?
NVIDIA states models up to 200 billion parameters at FP4 with 128GB of unified memory per unit.
Is DGX Spark a graphics card for my PC?
No. It is a self-contained computer with its own chip, memory and storage.