RTX Spark N1X laptops launch with local Nemotron agents

NVIDIA’s RTX Spark Windows laptops launched Oct. 16 on Arm N1X chips with up to 128GB memory and local Nemotron agents.

NVIDIA says RTX Spark Windows PCs for gaming, creative work, and local AI became available Oct. 16, 2026, with launch events that day in San Francisco, Berlin, Taipei, Beijing, Istanbul, and Seoul, and systems offering up to 128GB of unified memory and 1 petaflop of AI compute. The Verge reports the first laptops use the Arm-based N1X—20 CPU cores, a Blackwell GPU with 6,144 CUDA cores, and unified memory—rather than AMD or Intel x64 chips, and that Microsoft tuned Windows on Arm with a Store Get Started flow, Prism emulation, unified-memory management, and a GPU-memory slider. Sources differ on agents: NVIDIA lists GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy as being optimized, while The Verge says Hermes, Perplexity, and OpenClaw are the first supported Store agents and Meta’s Muse is expected later. On Oct. 15 NVIDIA released open Nemotron 3.5 Super VL, a 120-billion-parameter vision-language model it says matches its largest reasoning model’s accuracy at about one-fifth the size and delivers up to 5x peer throughput, plus open weights, datasets, and training recipes. Bolt 74B-A9B is a 2-bit 24GB model with 9B active parameters, while a 4GB Bolt E4B remains due Oct. 23—first described for game characters and later as a conversational agent for GeForce RTX, Jetson, and Spark. The Verge’s week with Surface Laptop Ultra, Asus ProArt P16, Lenovo Yoga 9n, and HP OmniBook Ultra machines priced about $2,600 to $6,999 found Windows on Arm apps generally usable and gaming often near 60fps at medium settings with DLSS, though CPU-bound titles lagged.

• RTX Spark Windows PCs became available Oct. 16, 2026, with launch events that day in San Francisco, Berlin, Taipei, Beijing, Istanbul, and Seoul.
• The Verge says the Arm-based N1X pairs 20 CPU cores with a Blackwell GPU of 6,144 CUDA cores and up to 128GB unified memory; NVIDIA also claims 1 petaflop of AI compute.
• Tested configs included Microsoft Surface Laptop Ultra, Asus ProArt P16, Lenovo Yoga 9n, and HP OmniBook Ultra, priced about $2,600 to $6,999.
• Apps were generally usable natively or via Prism; gaming often held near 60fps at medium settings with DLSS, while CPU-bound titles lagged.
• Agent lists differ: NVIDIA cites GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy being optimized; The Verge says Hermes, Perplexity, and OpenClaw are the first Store agents, with Meta’s Muse later.
• On Oct. 15 NVIDIA released open Nemotron 3.5 Super VL, a 120B vision-language model it says matches its largest reasoning model at about one-fifth the size and up to 5x peer throughput, with weights, datasets, and recipes.
• Bolt 74B-A9B is a 2-bit 24GB MoE with 9B active parameters; 4GB Bolt E4B remains due Oct. 23, first described for game characters and later as a conversational agent.
• Microsoft tuned Prism and unified memory, including a GPU-memory slider; OpenShell and Execution Containers add sandboxing; several anti-cheat games launch now, with Call of Duty Ricochet targeted for 2027.