- Nvidia DGX Spark runs larger AI models locally using massive unified 128GB memory efficiently
- Native CUDA support makes Spark ideal for advanced AI workloads on desktops
- Its Arm CPU and Blackwell GPU combination avoids expensive professional graphics cards
The highly-anticipated Nvidia DGX Spark has finally arrived as a very small desktop system built around the GB10 Superchip.
It features a shared 128GB of LPDDR5X memory, a specification that immediately separates the system from typical desktops and even most compact workstations.
And according to an early review by Toms Hardware, the system delivers strong results only when its AI-oriented capabilities are fully used.
Hardware design and connectivity focus
The Spark’s hardware design relies on a single package that combines an Arm-based CPU with a Blackwell GPU.
This integration allows Nvidia to support larger local models without requiring professional-class graphics cards at extreme costs.
While Apple and AMD systems offer large shared memory configurations, they lack direct support for Nvidia’s software ecosystem, which continues to dominate many AI development workflows.
The physical design emphasizes density and airflow instead of visual flair or modular expansion.
At just over one liter in volume, measuring roughly 150 by 150 by 50 mm, the unit fits comfortably among any modern mini PC, but the similarities mostly end there.
Alongside a USB-C power input, the unit provides three USB-C 20Gbps ports with DisplayPort alternate mode, an HDMI 2.1a port, and a 10Gb Ethernet connection.
Most notably, it includes two QSFP ports driven by an onboard ConnectX-7 network interface capable of up to 200Gbps, allowing multiple units to be linked together for distributed computing experiments, a capability rarely associated with a mini PC.
The system runs DGX OS, a customized Ubuntu 24.04 LTS distribution aligned closely with Nvidia’s software stack.
It can operate as a locally attached computer with a monitor and keyboard, or as a headless system accessed remotely over a network.
Nvidia’s Sync utility simplifies remote access from Windows and macOS machines, allowing AI tools to run continuously in the background.
These usage patterns resemble how mobile workstations or shared compute nodes are accessed, rather than how everyday desktops are typically used.
The DGX Spark benefits from a unified 128GB memory pool with native CUDA support, a pairing that is uncommon in compact systems designed for local AI work.
This configuration allows larger models to run fully in memory, avoiding frequent data movement between system RAM and GPU memory, and as a result, some of the practical limits seen on discrete GPUs with smaller VRAM pools are reduced.
That same capability also introduces clear trade-offs. The entry price remains high relative to compact desktops, especially for users who do not run demanding AI workloads every day.
The system does not support Windows, which restricts software compatibility for users outside Linux-focused environments.
Its GPU is also unsuitable for gaming or general graphics tasks, reinforcing its narrow scope.
DGX Spark assumes that local AI experimentation is a primary and ongoing requirement but if this is not your priority, it lose practical value.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.