Nvidia said RTX Spark laptops are available from ASUS, Dell, HP, Lenovo, Microsoft and MSI starting October 16. The machines combine a Blackwell RTX GPU and a Grace CPU with up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance, with MediaTek co-designing the CPU. Nvidia positions the chip for running large language models and agents locally; the launch includes guided installers for GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy. The company also points to Nemotron 3.5 Bolt 74B-A9B, a 74-billion-parameter model tuned for tool use, plus an OpenShell runtime and Microsoft Execution Containers for permission controls. Nvidia says RTX Spark can also serve inference to other PCs on a home network through NVIDIA PAIR. The company claims support for 90GB+ 3D scenes and AAA games at 1440p above 100 frames per second.
Our takeThe spec that matters is not the petaflop; it is 128GB of unified memory plus a permissioned agent runtime shipping by default. That combination moves local agents from a demo to a supported PC workload, and the day-one app list shows the software side is being seeded. The open question is cost and battery life, which Nvidia's release does not address. Watch whether developers actually run frontier models locally or treat the NPU as a checkbox.