AI · SEMICONDUCTOR · FACT CHECK
SKD WIREMostly True
NVIDIA's DGX Spark desktop AI machine ships with up to 128GB of memory and delivers up to 1 petaflop of AI performance, specifications that match the company's official product pages and CES announcements — though the headline figures come with caveats that matter for buyers assessing the rise of local AI.
The 128GB memory figure represents the maximum configuration of the DGX Spark, and the 1 petaflop performance rating is a theoretical peak achieved with FP4 precision and sparsity applied. In practical workloads, real-world throughput will fall below this theoretical maximum. The product pages and CES announcements are preserved through web archives, keeping the published numbers accessible for reference.
The framing of DGX Spark as a machine that expands the era of local AI — running large AI models on a desktop rather than in the cloud — is an interpretive narrative rather than a technical specification. It is, however, a reasonable reading of the product's positioning: a 128GB-memory desktop-class system aimed at developers and researchers working outside data centers is consistent with a broader push toward on-device and on-premises AI development.
NVIDIA's official materials support the core numbers, but consumers should note that the 128GB figure describes the top configuration and the petaflop rating describes peak theoretical performance, not guaranteed real-world output.
The claim is largely accurate on the facts, with the omitted caveats about peak performance and maximum configuration tempering the picture — the verdict is Mostly True.
Verdict: Mostly True