Nvidia Unveils Personal AI Agent Devices
量子位 · wechat · 2026-07-12
At BW, Nvidia showcased two new devices powered by Grace CPU + Blackwell GPU, connected via NVLink-C2C, and equipped with 128GB unified memory: the consumer-facing RTXSpark laptop/desktop and the developer-focused DGXSpark desktop supercomputer.
RTXSpark: For Personal Agents and Creators
- Positioned as a "local personal agent" device, it can run 120B parameter models locally with a maximum context length of 1 million tokens.
- The accompanying OpenShell runtime allows users to restrict agent executable actions, keeping sensitive information local and desensitizing data before sending it to the cloud.
- It also emphasizes creative and gaming capabilities: handling heavy 3D scenes, supporting 12K 4:2:2 video decoding, and stabilizing AAA game framerates at 100FPS+ in 1440p.
- Multiple game publishers have expressed support for Arm-native games, and a live demo featured the Arm-native version of Naraka: Bladepoint.
DGXSpark: A Desktop AI Supercomputer for Developers
- Featuring the same 1 Petaflop peak computing power and 128GB unified memory, but designed for proof-of-concept, fine-tuning, and inference.
- Supports models with up to 200B parameters, and can handle even larger models when two devices are networked together.
- A live demo showcased a personal agent based on the Qwen 35B multimodal model: generating a webpage from a hand-drawn sketch locally in just tens of seconds.
Other Info
- Nvidia also launched the GeForce Collector's Card Series, featuring 14 free collectible cards highlighting historical milestones in GeForce history.
More from coding & agent
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22
- BUZZ launches as an open-source group chat layer for teams and agents — Scobleizer · 2026-07-22
- A Firecracker-based platform says it can host 6,000 AI agents on one 256 GB server — maritime_sh · 2026-07-22
- A better path to agent autonomy is running waves, finding friction, and iterating — JnBrymn · 2026-07-22