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: HKR-H/K/R all pass: NVIDIA frames RTX Spark for local agents and gives hard specs: 1 petaflops, 128GB, and up to 2x llama.cpp throughput. Vendor-blog framing keeps it in the low 78–84 band.