NVIDIA chip powers local AI workloads
NVIDIA has unveiled the RTX Spark, a “superchip” delivering up to 1 petaflop of AI compute to enable Windows PCs to run personal AI agents. The device combines 128 GB of unified memory with an NVIDIA Blackwell RTX GPU featuring 6,144 CUDA cores and fifth-generation Tensor Cores that provide FP4 precision. The GPU connects to a high-performance 20-core Grace CPU via the NVLink-C2 chip-to-chip interconnect.
NVIDIA collaborated with MediaTek on the custom CPU design, contributing to strong power efficiency, performance, and connectivity. NVIDIA also partnered with Microsoft to deliver a secure Windows platform for on-device agents, incorporating new Windows security primitives and the NVIDIA OpenShell runtime to safely run autonomous AI agents.
RTX Spark brings NVIDIA’s AI and graphics technologies to creators, developers, and gamers. It can run 120-billion-parameter language models, render large 3D scenes, and accelerate 12K video editing. For gaming, the platform supports ray tracing and NVIDIA DLSS technologies for enhanced visual quality and performance.
RTX Spark-based laptops and compact desktops will be available this fall from leading manufacturers.
The post NVIDIA chip powers local AI workloads appeared first on EDN.
