NVIDIA Launches RTX Spark Superchip for Local Agentic AI
NVIDIA released the RTX Spark superchip for Windows PCs to enable high-performance local AI agents and reduce reliance on cloud APIs.
NVIDIA announced the RTX Spark superchip for Windows PCs, a hardware architecture designed to enable local agentic AI. The chip integrates a Blackwell RTX GPU and a 20-core Arm-based Grace CPU, developed in collaboration with MediaTek, sharing up to 128GB of unified memory. This design allows the GPU to process large language models and massive 3D scenes locally, which improves user privacy and reduces dependence on cloud APIs.
Hardware partners including ASUS, Acer, Dell, HP, Lenovo, Gigabyte, Microsoft, and MSI are releasing laptops featuring the new chip. Over 100 software providers are optimizing for the platform. Adobe is rearchitecting Photoshop and Premiere Pro to utilize TensorRT and unified memory, while Blackmagic Design has developed a native Arm build for DaVinci Resolve Studio 21. The system supports AI agents such as Hermes Agent and OpenClaw, which can operate Windows applications and rewrite code to resolve compatibility issues.
Early performance tests indicate that while RTX Spark trails Apple's M5 chips in battery endurance and single-core CPU speed, it provides superior multicore performance and faster local AI compute speeds.
