NVIDIA rolls out local AI lineup at GTC Berlin: DGX Spark 64GB ships, 8-unit clusters run frontier models
NVIDIA Blog · rss · 2026-10-21
At GTC Berlin, NVIDIA announced a sweeping local-AI refresh: RTX Spark brings personal agents to Windows PCs with up to 128GB unified memory and 1 petaflop of AI compute; DGX Spark 64GB ships October 23 and clusters up to 8 units via ConnectX-7 for 1TB of unified memory capable of running frontier-class models like GLM 5.3. DGX Station offers 748GB coherent memory and 20 petaflops for 30+ users, with a Windows version arriving in December. NVIDIA also launched the Recursive Flint local coding model, Day-0 support for Nemotron, GLM, DeepSeek, Qwen, MiniMax and FLUX.3, plus llama.cpp/vLLM accelerations and three open-source robotics reference workflows. Jensen Huang demoed training a Hugging Face robot with agents running on RTX Spark.
More from Embodied
- Humanoid robot bends over a car to inspect a person lying underneath — CyberRobooo · 2026-10-21
- Hugging Face Launches SimReady Central: Open Sim-Ready 3D Assets for Robotics — Hugging Face Blog · 2026-10-21
- Musk: Neuralink's Blindsight implant writes to visual cortex, first low-res version coming soon — XFreeze · 2026-10-21
- NVIDIA GTC Berlin: Mercedes, Volvo adopt Hyperion as Halos SAFE and Alpamayo 2 Edge debut — nvidia · 2026-10-21
- NVIDIA expands Isaac robotics platform with Agile ONE SN reference humanoid — nvidia · 2026-10-21
- Skild AI hits $100M+ ARR and $14B valuation with $1.4B SoftBank-led round — xmercury_one · 2026-10-21