NVIDIA Packs 1 Petaflop AI Into Thin Laptops

NVIDIA's RTX Spark laptops, built on a single Blackwell GPU and Grace CPU superchip with up to 128GB of unified memory, are now sold by ASUS, Dell, HP, Lenovo, Microsoft and MSI. The systems deliver up to 1 petaflop of FP4 AI performance in premium Windows laptops, bringing the CUDA platform, RTX graphics and what NVIDIA calls a personal AI supercomputer into machines designed to travel. The launch comes with a set of software updates for local AI, gaming and creative work, including a new open speech model, a new Liquid AI model family, a new resource management control and a launch-day game.

Inside the Superchip

RTX Spark combines an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU in one superchip. Up to 128GB of unified memory sits on the same package, so creative projects and AI models draw from one memory pool. NVIDIA says this lets a laptop take on work that used to require a desktop workstation, even when unplugged. The design also enables slim hardware: RTX Spark laptops are as thin as 14mm and as light as three pounds, with all-day battery life. Buyers can choose among different screen sizes, colors, touchscreens, pen support and two-in-one convertibles.

For AI developers, the pitch is local development on the same CUDA platform used in the cloud, with a path to deploy across more than 150 million NVIDIA RTX AI PCs and NVIDIA GPUs in the cloud. For creators, RTX Spark can edit camera-native 4:2:2 video and render 3D scenes of up to 90GB. Adobe Photoshop and Premiere offer native RTX Spark support through beta releases. For gamers, hundreds of games run on the platform, with NVIDIA DLSS, ray tracing, Reflex and G-SYNC aimed at sharper visuals and smoother gameplay.

Nemotron 3.5 Transcribe: One Model for Recorded Audio

NVIDIA also introduced Nemotron 3.5 Transcribe, an open weight speech model that turns recorded conversations into usable information. Developers can replace a chain of separate models for transcription, speaker labels and translation with one promptable model that handles all three, and can also ask questions about the audio. It transcribes 40 languages, including seven Indic languages.

Developers steer the model with plain-language prompts. For a single recorded meeting, examples include “Transcribe this recording and label each speaker,” “Translate this audio into English,” and “List what decisions were made in this meeting.” Prompts can also include product terms and formatting conventions, which helps with specialized vocabulary in healthcare, financial services and customer service. Because the weights are open, developers choose where audio is processed, from a desktop to the cloud.

NVIDIA says the model ranks No. 1 on the Hugging Face Open ASR leaderboard and No. 2 among open weight models on Artificial Analysis. It achieves this with approximately 4.3 billion active parameters, 80% fewer than the No. 1 open weight model, which NVIDIA says makes high-quality transcription more practical on local hardware. Quantized versions are coming soon for local use on RTX Spark, DGX Spark systems, Jetson AGX Thor devices, or scaled up on NVIDIA Blackwell data center GPUs. NVFP4 versions will be deployable through NeMo-Speech.cpp or vLLM.

Agents and Apps Share the GPU

RTX Spark is designed to run long-running agents alongside everyday applications, and both share the system's GPU compute. The new resource management feature in the NVIDIA App lets users decide how that GPU time is divided. They can prioritize the foreground app for the best performance, or give a background agent more resources to finish work faster. The control is found under System > Advanced in the NVIDIA App and requires hardware-accelerated GPU scheduling to be enabled in Windows Settings.

Liquid AI's LFM3 Models

Liquid AI's LFM3 is a new generation of Liquid Foundation Models for specialized tasks. The family spans text, vision, audio, omni and decision models, built on one shared backbone, with base and post-trained models available as open weights. Liquid AI says its new 1.7-billion-parameter base model outperforms its previous 2.6-billion-parameter base model. Use cases include agents running in Hermes Agent, Pi and OpenClaw that triage inboxes and manage calendars, read documents and app screens, transcribe and translate speech, answer spoken questions about handwritten math, and route support tickets or voice requests. Developers can run LFM3 models on RTX Spark with native support through llama.cpp and vLLM, and fine-tune them locally.

Phantom Blade Zero Launches With Ray Tracing and DLSS 4.5

S-GAME's action role-playing game Phantom Blade Zero launches October 28 at 7 p.m. PT on Steam and the Epic Games Store. Players take on Soul, a swordsman framed for his master's murder, as he travels the martial world to uncover a conspiracy. NVIDIA worked with S-GAME Studio and S-GAME Publishing to add ray-traced effects. On RTX Spark laptops and GeForce RTX GPU-powered PCs, ray tracing adds more accurate bounced lighting, detailed reflections and realistic shadows on foliage and terrain. NVIDIA DLSS Dynamic Multi Frame Generation boosts frame rates with up to 6x frame generation, or adjusts the multiplier on the fly to hit a player's target frame rate.

RTX Remix 1.6 Adds Minecraft and DLSS 5

RTX Remix, NVIDIA's modding platform for remastering classic games with path tracing and DLSS, released version 1.6. The update brings RTX Remix to Minecraft: creators and players can render Minecraft Java Edition 1.21.1 through RTX Remix with DLSS-accelerated path tracing. Version 1.6 also adds DLSS 5 with 3D-Guided Neural Rendering, up to 6x Multi Frame Generation, native RTX Spark support and agent-assisted workflows. Portal with RTX and the Half-Life 2 RTX demo both now support RTX Spark, with DLSS 3D-Guided Neural Rendering for more lifelike lighting and materials, a new Performance Mode for DLSS Frame Generation and new options that speed up path tracing.

For modders, a redesigned ComfyUI workflow generates physically based materials directly inside RTX Remix, with no application programming interface setup. It adds batch processing for full projects and presets for less technical creators. NVIDIA agent skills also let compatible AI agents work directly with RTX Remix to speed up remastering.

More Local AI Updates

• Microduck robot: At the GTC Berlin keynote, Jensen Huang showed AI agents on RTX Spark helping train Hugging Face's Microduck robot. Hermes Agent and Nemotron 3.5 Bolt plan an obstacle course, Astra builds the 3D assets, NVIDIA Isaac Lab trains the navigation policy, and RTX Spark validates it in NVIDIA Isaac Sim.
• Nemotron 3.5 Bolt E4B: This model is now available and enables natural conversations and multistep agent tasks with as little as 4GB of memory. It powers local voice assistants, game characters and embedded interfaces on RTX Spark, DGX Spark, GeForce RTX GPUs and NVIDIA Jetson devices. It supports llama.cpp, vLLM, Ollama, Unsloth and LM Studio, and works with Hermes and OpenClaw. For heavier agent workloads, Nemotron 3.5 Bolt 74B-A9B 24GB is designed to run local agents on RTX Spark.
• FLUX 3 Image Klein: Black Forest Labs' 5.6-billion-parameter open weights model generates and edits images locally. NVIDIA-optimized versions run on RTX GPUs with as little as 8GB of VRAM, and developers can fine-tune it on RTX Spark using an NVIDIA NeMo AutoModel recipe.
• DGX Spark 64GB: Now available to purchase from manufacturer partners, this system is powered by the GB10 Grace Blackwell Superchip and combines 64GB of unified memory with DGX OS and the NVIDIA AI software stack for agent development, inference and fine-tuning.

Key Features at a Glance

• Superchip design: Blackwell RTX GPU and Grace CPU in one package, with up to 128GB of unified memory, so AI models and creative projects share a single memory pool.
• AI performance: Up to 1 petaflop of FP4 AI performance, for running models and agents locally on a laptop.
• Portable build: As thin as 14mm and as light as three pounds, with all-day battery life.
• Nemotron 3.5 Transcribe: One open weight model for transcription, speaker labels, translation and audio Q&A across 40 languages, using about 4.3 billion active parameters.
• Resource management: Users choose whether the foreground app or a background agent gets more GPU resources, set under System > Advanced in the NVIDIA App.
• Creative tools: Edits camera-native 4:2:2 video and renders 3D scenes of up to 90GB, with Adobe Photoshop and Premiere beta support.
• Gaming: DLSS, ray tracing, Reflex and G-SYNC, with Phantom Blade Zero launching with DLSS 4.5 and ray-traced lighting.

Availability

RTX Spark laptops are available from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599, with 128GB configurations on shelves. Find a laptop on the RTX Spark product page. Nemotron 3.5 Transcribe is available on Hugging Face, with quantized versions coming soon. Phantom Blade Zero launches October 28 at 7 p.m. PT on Steam and the Epic Games Store. NVIDIA DGX Spark 64GB is available to purchase from manufacturer partners.