Nvidia launches RTX Spark laptops with 128 GB unified memory and one petaflop of AI compute

Starting Friday, October 16, the first laptops built on RTX Spark hit shelves — a single chip that merges a gaming workstation, a creative studio and a personal AI supercomputer. The spec sheet lists up to 128 GB of unified memory and one petaflop of AI performance, letting the full RTX and CUDA stack run locally without cloud dependence. Nvidia and Microsoft announced availability last week alongside a wave of software updates and optimizations designed to expand what local AI agents can do on the new platform.

Local agents: apps, models, security

On the agent side, Nvidia is partnering with a roster of leading applications — GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy — which are receiving updates for faster installation on RTX Spark, hardened security via Nvidia OpenShell and Microsoft's MXC execution containers, and hybrid routing between local and cloud processing. Nemotron 3.5 Bolt 74B, a model purpose-built for agentic AI on the platform, releases tomorrow, October 15. In parallel, Nvidia's PAIR tool exits beta with added llama.cpp support, simplifying local deployment of open models.

Creative apps: Adobe, Broadcast, ComfyUI, Blender

Nvidia and Adobe have re-architected Photoshop for GPU-accelerated compositing and Premiere for playback, rendering and effects, promising up to two times performance gains in upcoming updates. Nvidia Broadcast 2.3 brings Super Resolution to video, frame generation, offline file processing and agent control via the MCP protocol. ComfyUI already offers DLSS frame generation and HDR capabilities to accelerate generative video workflows, while Blender 5.3 will receive DLSS Ray Reconstruction in November to improve viewport interactivity.

Developer and gaming tools: CUDA, DLSS 5, ACE

For developers, every major framework — PyTorch, Python, CUDA and the full CUDA library suite — is ready for RTX Spark, and both GitHub Copilot and Pi have been optimized. llama.cpp adds decision-model support and launches a dedicated Windows app. On the gaming front, hundreds of top titles and anti-cheat systems are supported; a DLSS 5 model twice as fast arrives at launch, and a new Performance mode for DLSS Frame Generation is already available in the Nvidia App beta. The compact Nemotron 3.5 Bolt E4B 4GB language model for ACE AI characters releases October 23, accompanied by an RTX Remix update that includes DLSS 5 support.

The local-agent surge: five-fold download growth in a year

The launch rides a sharp uptick in local AI agent adoption: projects such as OpenClaw and Hermes lead GitHub popularity charts, and local models like Qwen 3.8-Flash-Next are closing the gap with advanced cloud models released only months ago. The result — downloads of local LLMs running five times higher than all of last year. On first boot, RTX Spark presents a "Quick Start" screen for selecting and installing an agent application, with an OS-level sandbox via MXC, OpenShell access and permission guardrails.