RTX Spark and MXC on Windows

RTX Spark laptops go on sale October 16, combining a Blackwell RTX GPU in 5,120-core or 6,144-core configurations with a Grace CPU connected at 600 GB/s, one petaflop of FP4 AI performance, and up to 128GB of LPDDR5x unified memory. Microsoft's Surface Laptop Ultra starts at $2,599 and the Surface RTX Spark Dev Box at $5,999; OEM partners include Acer, ASUS, Dell, HP, Lenovo, MSI, and Gigabyte, with compact desktops following in November. NVIDIA described the chip as combining gaming PC, creative studio, and personal AI supercomputer capabilities in a single chip.

Microsoft Execution Containers (MXC) are generally available on Windows 11, providing OS-level sandboxing for agent workloads with per-agent file and network access policies; isolation levels range from separate processes and sessions up to virtual machines and cloud PCs. HydraFusion enables models to run locally while accessing the cloud for additional scale, with GitHub collaboration in progress. Local agent apps including Muse with MXC integration, Hermes, OpenClaw, and Perplexity are being built for Windows, and Windows taskbar Search is adding thousands of quick actions this fall.

Microsoft demonstrated DeepSeek V4 Flash, a 284B-parameter open-weight model, running locally on a Surface Laptop Ultra after being quantized to 1.6 bits, which reduces its memory requirement to approximately 60GB. Pavan Davuluri of Microsoft said the Surface Laptop Ultra was built around RTX Spark and that users 'can run models on this laptop that simply don't fit on a traditional machine.'

NVIDIA releases Nemotron 3.5 Bolt 74B on October 15, a 74B-parameter MoE model with 9B active parameters compressed to 24GB via 2-bit quantization; NVIDIA describes it as aiming to be 'faster than most popular dense models while smarter than most popular MoE models with a similar memory footprint.' The model is fully open source and available through llama.cpp, vLLM, Ollama, Unsloth, LM Studio, Hermes, and OpenClaw.

Adobe is collaborating with NVIDIA to deliver up to 2x performance improvements in Photoshop, Premiere, and Substance 3D for RTX Spark, with GPU-accelerated compositing and AI feature optimizations available in beta. The first native PyTorch-CUDA build for Windows on Arm, Python 3.15, and a native Conda installer with over 2,000 packages are releasing alongside the hardware; 97% of the top 15,000 Python wheels are ready at launch.

Llama.cpp is adding decision model support, which NVIDIA describes as allowing agents to classify requests and select next actions up to 200x faster than traditional reasoning loops, and is releasing Llama Windows, a native Windows 11 system-tray app for running local models. Microsoft is adding llama.cpp as a backend to Windows ML.

DLSS 5 with 3D-Guided Neural Rendering launches on RTX Spark October 16, with a new 2x faster AI model debuting in NBA 2K27 that boosts frame rates by up to 1.4x; Hogwarts Legacy follows November 5. On October 23, NVIDIA releases RTX Remix 1.6, which replaces its REST API node-based texture workflow with ComfyUI Version 2 embedded in the Remix sidebar, enabling PBR material generation with no API setup required. Also on October 23, Nemotron 3.5 Bolt E4B 4GB launches for in-game AI characters; SkyrimNET uses it to make every Skyrim NPC an 'unscripted, interactive character equipped with personal memory, voice and autonomy.'

For enterprise, NVIDIA previewed DGX Station for Windows, powered by the GB300 Grace Blackwell Ultra Desktop Superchip, delivering 748GB of coherent memory and up to 20 petaFLOPS of FP4 compute. Previously DGX Station ran Linux only; NVIDIA noted the vast majority of Fortune 500 companies are standardized on Windows. RTX Spark runs the full NVIDIA CUDA platform, enabling developers to move workloads from RTX Spark laptops to DGX Station without rewriting code.

NVIDIA stated that local LLM downloads are already 5x higher than all of last year, driven by agent popularity.