NVIDIA announced on October 16 that RTX Spark laptops are officially available, with Asus, Dell, HP, Lenovo, Microsoft, and MSI as the six launch partners, and more models to follow. This is the first time Blackwell RTX graphics, CUDA, and the full-stack AI platform have been packed into thin-and-light Windows laptops, with up to 128GB unified memory and 1 petaflop of FP4 AI performance. Taiwan suggested retail price starts at NT$119,069.
A superchip: Blackwell plus Grace, with MediaTek handling the CPU
At the heart of RTX Spark is a superchip: a Blackwell RTX GPU paired with a Grace CPU, with up to 6,144 CUDA cores and fifth-generation Tensor cores on the GPU and up to 20 cores on the CPU. The two chips are connected via NVLink-C2C and run Windows 11. The custom CPU was co-designed by MediaTek and NVIDIA and is touted for best-in-class energy efficiency, performance, and connectivity. The shared memory architecture lets heavy creative projects and local AI models run simultaneously. CUDA, DLSS, OptiX, Reflex, G-SYNC, and TensorRT are all packed into a chassis that fits in a shoulder bag, and the battery is claimed to last all day.
From Tools to Teammates: Agents Don’t Need to Type Commands
NVIDIA’s slogan this time is that the PC goes from being a tool to a partner working side by side with you. Guided setup plus a simple installer means you can run GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy without touching the command line. Dillon Rolnick, CEO of Hermes developer Nous Research, said that bringing Hermes to RTX Spark lets users run open models and agents locally on Windows laptops, giving them greater control over their data and the ability to shape AI around their own workflows. NVIDIA’s own Nemotron 3.5 Bolt 74B-A9B is also joining the lineup, with 74 billion parameters tuned specifically for tool calling and multi-step tasks.[1]Access control runs through the OpenShell secure execution environment plus Microsoft Execution Containers, while agent skills can call Broadcast, RTX Remix, and Omniverse libraries—for example, adjusting effects during a livestream or batch-processing prerecorded clips, all done locally. NVIDIA PAIR also lets RTX Spark provide AI inference services to Windows, Linux, and macOS computers on a home network.
Creation and development: native Adobe support, CUDA stays local
At launch, Adobe is using TensorRT to accelerate AI editing features in Photoshop, Premiere, and Substance 3D; Photoshop and Premiere offer native support in beta at launch. Broadcast 2.3 can remove background noise from recordings, the beta version of RTX Video offers up to 4K super resolution plus 4x frame generation, and ComfyUI’s new node can turn 15fps AI video into 60fps HDR without cloud inference fees. On the developer side, native CUDA lets models and agents run locally, keeping code and context on-device, and they can later be deployed to more than 150 million RTX AI PCs or cloud GPUs. NVIDIA and Microsoft have prepared Windows on Arm’s first PyTorch-CUDA build, first Python that runs natively by default, and first native Conda installer, supporting 97% of the 15,000 most popular Python packages.
Games: DLSS 5 available on day one, with hundreds of titles ready to play
The game lineup is a launch highlight. Cyberpunk 2077, Fortnite, Gears of War: E-Day, Marvel Rivals, World of Warcraft, and NBA 2K27 all launch on day one, with anti-cheat configured for online games and full support for DLSS, Reflex, G-SYNC, and ray tracing. NBA 2K27 directly uses DLSS 5’s 3D-guided neural rendering, exceeding 100fps at 1440p and capable of rendering 3D scenes over 90GB. The day-one Game Ready 617.42 WHQL driver launches at the same time.
Launch Information
Launch models from six brands: 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. Available starting today on global official websites and major retail channels, starting at US$2,599 in the US and NT$119,069 in Taiwan. Limited-time offer for new Perplexity users; prices and configurations vary by region.
Source: KOCPC Chinese