NVIDIA Releases RTX Spark and Local AI Agent Security and Performance Updates
NVIDIA 发布 RTX Spark 及本地 AI 智能体安全与性能更新
NVIDIA released RTX Spark, a Windows PC for local AI agents with 1 petaflops of AI compute and 128GB of unified memory. OpenShell uses new Windows security primitives with Microsoft, while llama.cpp optimizations raise Qwen 27B throughput by up to 2x.
Why it matters: At Computex Nvidia showed RTX Spark, which runs large models locally with 1 petaflops and 128GB unified memory, plus an OpenShell runtime built with Microsoft to keep agents on-device.