NVIDIA RTX Spark laptops with local AI agents are now available worldwide

NVIDIA announced that RTX Spark laptops went on sale worldwide starting October 16. Models from ASUS, Dell, HP, Lenovo, Microsoft, and MSI are being sold through the manufacturers’ own stores and major retailers, with a starting price of $2,599. Turkey is also on the launch map: NVIDIA is holding live demos in Taipei, Seoul, Beijing, Istanbul, San Francisco, and Berlin on the same day.

Models going on sale include the ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI Prestige N16 Flip AI+ and Prestige N16. The systems support local AI agents, creative apps, and games from day one.

Blackwell GPU and Grace CPU on a single chip

Every RTX Spark system is built on a single superchip that combines a Blackwell RTX GPU and Grace CPU with up to 128 GB of unified memory on the same chip. NVIDIA says the platform delivers up to 1 petaflop of FP4 AI performance, enough to run large models and agents directly on the device. The CPU design was developed in collaboration with MediaTek.

There are two laptop configurations: a 6,144-core GPU, 20-core CPU, and 128 GB LPDDR5X memory; or a 5,120-core GPU, 18-core CPU, and 64 GB memory. Both run Windows 11 in a 45-80 W power range and support fourth-generation ray tracing cores, fifth-generation Tensor cores, DLSS 5, and Reflex 2.

AI agents out of the box

Setup wizards install agents such as GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy without requiring a command line. Nous Research CEO Dillon Rolnick says they brought the Hermes agent to RTX Spark and users can run open models locally on a Windows laptop. The 74-billion-parameter Nemotron 3.5 Bolt 74B-A9B model, built for tool calling and multi-step tasks, can also be used on supported configurations.

NVIDIA OpenShell secure sandbox and Microsoft Execution Containers are in place to manage agent permissions. NVIDIA agent skills allow compatible agents to work with NVIDIA Broadcast, RTX Remix, and Omniverse libraries. NVIDIA PAIR lets a single RTX Spark computer perform AI inference for compatible Windows, Linux, and macOS machines on the same home network.

What’s in it for developers and creators

Adobe is adding TensorRT acceleration to the AI editing features in Photoshop, Premiere, and Substance 3D; Photoshop and Premiere support arrives in beta at launch. NVIDIA Broadcast 2.3 cleans up background noise in existing recordings and adds Studio Voice microphone profiles. Beta RTX Video features offer up to 4K super-resolution and up to 4x frame generation. New RTX Video nodes in ComfyUI can convert 15 fps AI videos into 60 fps HDR video without per-use cloud fees.

NVIDIA says the systems can edit 8K 4:2:2 video on battery and handle 3D scenes up to 90 GB. On the developer side, the first PyTorch-CUDA build, first native Python build, and first Conda installer for Windows on Arm are ready; 97% of the 15,000 most popular Python packages are supported.

On the gaming side

At launch, Cyberpunk 2077, Fortnite, Gears of War: E-Day, Marvel Rivals, World of Warcraft, and NBA 2K27 are supported. NBA 2K27 uses DLSS 5’s 3D-enabled neural rendering feature. NVIDIA says a new AI model that went live on October 16 and is, in its words, twice as fast boosts DLSS 5 frame rates by up to 1.4x. According to the company, RTX Spark can run AAA games at over 100 fps at 1440p.

NVIDIA and Microsoft first announced RTX Spark at the Windows AI and Surface event in San Francisco on October 7, and preorders opened with an October 16 delivery date. Compact desktop configurations are planned to go on sale in November.

Source: Unite.AI