Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.
Nvidia’s DGX Spark is a bit tough to judge at first. It isn’t meant to replace your existing PC or Mac, although it certainly could if you’re a Linux native. It's not built to be a gaming system, although it can certainly do it as a party trick.
Instead, the Spark is best understood as an AI development toolbox, a Swiss Army knife for developers, institutions, and curious local AI enthusiasts who want a stable, dependable platform to build and explore with.
The Spark’s 128GB of RAM and the solid performance afforded by its Blackwell GPU across a wide range of workloads mean that you can work without getting stopped cold by resource limitations most of the time, and for a jack-of-all-trades development sandbox, that’s mission accomplished.
Beyond the hardware, time is money in a field that’s moving this fast, and Nvidia’s thorough documentation and software support are incredibly valuable if you want to get local AI workloads up and running without spending hours digging through Reddit threads or GitHub pages to figure stuff out.
The Spark is also interesting purely as a milestone in personal computing. It’s the first “real” PC built entirely under Nvidia’s direction. Here’s a fast, efficient SoC on a cutting-edge 3nm process, boasting a 20-core Arm CPU paired with a powerful Blackwell GPU using the NVLink C2C interconnect, and it all just works. There’s not a drop of AMD or Intel tech in this platform, and maybe that should keep some folks up at night.
The $4,000 price tag of the Founders Edition Spark we tested might seem high on its face, but given its performance and versatility, it’s plenty competitive with other products in the market right now.
If you just want a powerful local token generator with lots of RAM for the biggest models, the comparison for the DGX Spark is probably a system like Apple’s Mac Studio.
The Studio’s much higher memory bandwidth (546 GB/s vs the Spark's 273 GB/s) should result in higher tokens-per-second throughput from LLMs if that’s your sole point of interest.
You can configure an M4 Max-powered Studio with comparable amounts of RAM and storage, along with a 40-core GPU, for about $700 more than the Spark as of this writing. But the Mac Studio doesn't support CUDA, and that will be a deal-breaker for some.
AMD’s Ryzen AI Max+ 395, aka Strix Halo, platform might look comparable to the DGX Spark on paper thanks to its powerful Radeon 8060S GPU and RAM pools ranging up to 128GB. Its x86 CPU makes it great for Windows gaming in addition to AI and Linux tasks. It’s developed a reputation as the local AI enthusiast’s scrappy platform of choice.
In our experience, though, AMD’s ROCm framework still isn’t as mature or stable as CUDA despite recent improvements (at least under Linux), and the Radeon 8060S has about half the raw GPU horsepower as the Spark. Our testing revealed that its performance at long LLM context lengths and for generative workflows falls well short of GB10’s.
If you need to tinker with large AI models or agentic workflows for the lowest possible cost, Strix Halo used to be a relatively inexpensive way of getting into the game, but prices for these systems have risen sharply amid the current AI crunch. You can still configure a wide range of Ryzen AI Max+ 395 boxes with 128GB of RAM and 4TB of storage for about $3,000 from vendors like Corsair, Framework, Minisforum , and GMKtec as of this writing.
But Asus’ Ascent GX10 version of the Nvidia GB10 platform also sells for as little as $3,000 with a 1TB SSD right now, and after our testing, we’d take the Ascent GX10 any day for a general-purpose local AI system. You can drop your own 4TB M.2 2230 SSD into the GX10 and keep about $700 in your pocket versus the Founders Edition Spark we tested.
AMD is going all in on Strix Halo for local AI this year, even going so far as to create its own Ryzen AI Halo mini-PC design that’s essentially a carbon copy of the DGX Spark. But hardware is only one part of the package, and from our experience, the Spark delivers the whole enchilada right now.
As local AI workflows continue to evolve rapidly, we think the DGX Spark (and the GB10 ecosystem more broadly) is the platform you’ll want to fully explore everything that’s possible from the whirlwind of new models and applications that will surely come down the pipe as we get further into 2026.
MORE: Best Graphics Cards
MORE: GPU Benchmarks and Hierarchy
MORE: All Graphics Content
