Nvidia debuts RTX Spark superchip for on-device AI
RTX Spark offers 1 petaflop and up to 128GB memory to run large AI agents locally on Windows laptops and desktops, with Microsoft-backed security and fall availability.
Nvidia introduced the RTX Spark superchip to run large AI agents locally on Windows laptops and desktops. The company said systems using the chip will be available from major PC makers this fall.
The RTX Spark pairs a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores with a 20-core Nvidia Grace CPU. Nvidia lists one petaflop of performance and up to 128GB of system memory on the module to support on-device AI inference.
Nvidia said the chip can run 120-billion-parameter language models with context windows up to one million tokens, enabling longer conversations and larger local model workloads without sending raw data to cloud servers. The company highlighted uses ranging from writing and creative workflows to gaming and video editing and noted mobile designs target a full day of battery life.
The processor was developed alongside MediaTek and under a three-year partnership with Microsoft. Nvidia and Microsoft added new core protections in Windows that aim to keep on-device agents under user control and reduce the risk of exposing personal information. The software layers include features to mask personal data before sending queries to cloud models, and Nvidia reported several agent developers have adopted the new Windows security primitives.
PC makers expected to ship RTX Spark systems this fall include Asus, Dell, HP, Lenovo and MSI, with Microsoft’s Surface line, Acer and Gigabyte indicating plans for Spark-based designs. Nvidia said the chip will target both laptops and desktops and that it intends to support existing developer tools and agent frameworks for on-device deployment.
Nvidia did not disclose pricing for RTX Spark-equipped systems. The company described the product as a platform combining hardware, software and partnerships to enable private agent experiences on Windows.
Nvidia founder and CEO Jensen Huang stated, “With RTX Spark and Microsoft Windows, you ask — and the PC does the work.” Microsoft chairman and CEO Satya Nadella described the company’s aim as delivering “unmetered intelligence to every home and every desk with Windows.” Vincent Koc, chief architect at the OpenClaw Foundation, noted that running OpenShell and Microsoft’s security primitives on RTX Spark will enable a fully integrated stack for private, personal agents on device.
Articles by this author
