Nvidia DGX Spark review: the GB10 Superchip powers a fast and fun AI toolbox that beats out AMD’s Ryzen AI Max+ 395

The fruits of the AI gold rush thus far have frequently been safeguarded in proprietary frontier models running in massive, remote, interconnected data centers. But as more and more open models with state-of-the-art capabilities are distilled into sizes that can fit into the VRAM of a single GPU, a burgeoning community of local AI enthusiasts has been exploring what’s possible outside the walled gardens of Anthropic, OpenAI, and the like.

Today's hardware hasn’t entirely caught up to the rapid shift in resources that AI trailblazers demand, though. Thin and light x86-powered "AI PCs” largely constitute familiar x86 CPUs with lightweight GPUs and some type of NPU bolted on to accelerate machine learning features like background blur and replacement. These systems rarely come with more than 32GB of RAM, and their relatively anemic integrated GPUs aren’t going to churn through inference at the rates enthusiasts and developers expect.

Gaming GPUs bring much more raw compute power to the table, but they still aren't well-suited to running more demanding local models, especially large language models. Even the RTX 5090 "only" has 32GB of memory on board, and it’s trivial to exhaust that pool with cutting-edge LLMs. Getting the model into RAM is just part of the problem, too. As conversations lengthen and context lengths grow, the pressure on VRAM only increases.

If you want to get more VRAM on a discrete GPU to hold larger AI models, or more of them at once, you're looking at professional products like an $8500+ RTX Pro 6000 Blackwell and its 96GB of GDDR7 (or two, or three, or four). And that’s not even counting the cost of the exotic host system you’ll need for such a setup. (Can I interest you in a Tinybox for $60K?)

The hunger for RAM extends to other common AI development tasks, like fine-tuning an already trained AI model for better performance on domain-specific data or quantizing an existing model to reduce its resource footprint for less powerful systems.

In the face of this endless hunger for RAM, systems with large pools of unified memory have become attractive platforms for those looking to explore the frontiers of local AI. Apple paved the way with its M-series SoCs, which in their latest and greatest forms pair as much as 512GB of LPDDR5 and copious memory bandwidth with powerful GPUs.

And AMD’s Ryzen AI Max+ 395 (aka Strix Halo) platform has found a niche as a somewhat affordable way to get 128GB of RAM alongside a relatively powerful GPU for local LLM tinkering.

But none of those systems natively support Nvidia’s CUDA, which remains the dominant software platform for AI development the world over. Enter the DGX Spark, first announced all the way back at CES 2025, which brings the combo of a high-performance Arm CPU and a Blackwell GPU to desktops for the first time alongside full support for the CUDA ecosystem.

The centerpiece of the DGX Spark is Nvidia’s GB10 SoC, which combines a MediaTek-produced Arm CPU complex joined together with a Blackwell GPU on one package. Both of these chiplets are fabricated on a TSMC 3nm-class node, and they’re joined together by Nvidia’s coherent, high-bandwidth NVLink C2C interconnect. Those chips share a coherent 128GB pool of LPDDR5X memory that offers a tantalizing canvas for pretty much any common AI workload you can think of.

Ports, external stuff, operating system and software

The Spark itself is a pretty straightforward mini PC. It measures just 5.9” by 5.9” by 2” (150mm by 150mm by 50.5mm) for a volume of just 1.1 liters.

The Founders Edition version we’re reviewing today has a spiffy gold finish with a metal foam front and back panel for ventilation. The “rack handles” at the front of the device conceal air intakes, but the top and sides are otherwise featureless.

Flip the Spark over, and you see another air intake along the bottom edge, as well as a removable rubber foot that conceals the wireless networking antennas and allows for access to the user-replaceable M.2 2242 SSD. Our unit came with a 4TB drive.

Around back, you get a power button, one USB-C power input, three USB-C 20Gbps ports with DisplayPort alt mode support, an HDMI 2.1a port, a 10Gb Ethernet port, and two QSFP ports for the onboard ConnectX 7 NIC running at up to 200 Gbps. That exotic NIC lets you cluster a pair of Sparks together for experimentation with Nvidia’s NCCL distributed computing libraries.

If the Founders Edition Spark isn’t to your liking, Nvidia has made the GB10 platform available to its system partners with a bit of wiggle room for customization. Dell, Acer, Asus, Gigabyte, HP, Lenovo, and MSI have all created GB10 boxes of their own with small variations in power, cooling, storage, cosmetics, and remote management options. Those options are likely of most interest to corporate and institutional IT departments that already have preferred vendors and support contracts.

Nvidia makes it easy to integrate the Spark into your existing workflow in a number of ways. The preinstalled DGX OS is a lightly Nvidia-flavored version of Ubuntu 24.04 LTS. You can use it as a regular PC with keyboard, mouse, and monitor connected, or you can set it up headless and use the free Nvidia Sync app to SSH into the system from your Windows PC or Mac. Using tools like Tailscale, you can conceivably connect to your Spark from anywhere in the world.

The included Nvidia Sync utility for Windows and macOS is handy, and it’s easy to set up. For just one example, I was able to leave ComfyUI running on the Spark and created a custom port forwarding rule that let me turn any PC in my house into a generative AI workstation with a single click. You can also set up Ollama for a private web chat interface or Cursor for AI-assisted coding in the same way.

MORE: Best Graphics Cards

MORE: GPU Benchmarks and Hierarchy

MORE: All Graphics Content