NVIDIA RTX Spark Delivers 1 Petaflop AI Performance and 128GB Memory
NVDA•NVIDIA unveiled RTX Spark superchip with 1 petaflop AI performance, 6,144 CUDA cores and 128GB unified memory powering PCs for local AI agents. Partnership with Microsoft delivers Windows security primitives and OpenShell to run 120B-parameter models, render 90GB 3D scenes and edit 12K video on OEM laptops and desktops.
1. Product Launch
NVIDIA introduced the RTX Spark superchip as its first Windows-native AI processor built for personal agents. The launch marks a strategic shift from traditional app-driven PCs to systems that natively run AI workloads and local agents with conversational interfaces.
2. Technical Specifications
The RTX Spark superchip combines a Blackwell RTX GPU with 6,144 CUDA cores, fifth-generation FP4 Tensor Cores and an NVLink-connected 20-core Grace CPU, delivering 1 petaflop of AI compute and up to 128GB of unified memory for demanding tasks.
3. Microsoft Collaboration
In collaboration with Microsoft, NVIDIA integrates new Windows security primitives and its OpenShell runtime to ensure on-device agents run under strict identity, containment and policy controls. This stack supports 120-billion-parameter models, enhances privacy routing and enables developers to build secure AI-powered Windows apps.
4. OEM Availability
RTX Spark will power slim, all-day battery Windows laptops and compact desktops from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI this fall, with models from Acer and GIGABYTE following. Systems will support ultralarge 90GB 3D scene rendering, 12K video editing and AAA gaming at over 100 FPS.


