TREE NEWS update: Nvidia said compact desktop PCs built on its RTX Spark platform are now on sale, with Asus, Dell and MSI already shipping models and Acer, Gigabyte and HP to follow. The machines pair a 20-core Nvidia Grace CPU with a Blackwell-based RTX GPU, up to 128GB of unified memory and up to 1 Petaflop of AI compute, enough to run 120-billion-parameter large language models locally. Nvidia said local inference cuts reliance on cloud APIs and token costs.
Nvidia RTX Spark Mini AI Workstations Launch, Run 120B-Parameter Models Locally
The pitch here is less about raw performance than about shifting where inference happens. Running 120B-parameter models on a desk moves a slice of AI compute away from cloud providers and onto hardware Nvidia sells directly — a positioning that matters to enterprises weighing token costs against capital outlay, and to cloud API economics more broadly. The breadth of OEM support, with more vendors still to ship, suggests Nvidia is treating local inference as a durable product category rather than a niche. Whether that translates into real enterprise adoption, and how cloud providers respond, is the open question.
Generated by AI for reference only.
Share on WeChat
Open WeChat → Scan → then tap "…" to send to a chat or Moments.
Tap "…" in the top-right corner to send to a chat or share to Moments.