NVIDIA says RTX Spark laptops are now available from ASUS, Dell, HP, Lenovo, Microsoft and MSI, bringing Blackwell RTX graphics, Grace CPUs, up to 128 GB of unified memory and up to 1 petaflops of FP4 AI performance to thin Windows PCs. The company also announced upcoming local-AI software, gaming and creator features, including Nemotron 3.5 Transcribe, Liquid AI’s LFM3 models, new NVIDIA app GPU controls, Phantom Blade Zero with ray tracing and DLSS 4.5, and RTX Remix 1.6 support for Minecraft.
RTX Spark combines AI performance and portability
RTX Spark laptops combine NVIDIA Blackwell RTX GPUs and NVIDIA Grace CPUs in a single superchip. With up to 128 GB of unified memory and up to 1 petaflops of FP4 AI performance, the systems are designed for developers, creators and gamers who need local CUDA, RTX graphics and AI capabilities in a portable Windows PC.
New RTX Spark features coming soon include:
- NVIDIA Nemotron 3.5 Transcribe, an open-weight audio model for transcription, speaker attribution and translation.
- Day-zero support for Liquid AI’s open LFM3 model family.
- NVIDIA app resource-management controls for allocating GPU resources between foreground applications and background agents.
- Phantom Blade Zero, S-GAME’s action role-playing game, launching with ray tracing and NVIDIA DLSS 4.5.
- Minecraft support for path tracing and DLSS through RTX Remix 1.6.
NVIDIA Nemotron 3.5 Transcribe processes audio locally
NVIDIA has introduced NVIDIA Nemotron 3.5 Transcribe, an open-weight audio model designed to turn recorded conversations into useful information. Instead of requiring separate models for transcription, speaker labels and translation, developers can use one promptable model for all three tasks, as well as for answering questions about audio.
The model supports 40 languages, including seven Indian languages. Users can guide it with plain-language prompts such as:
- “Please transcribe this recording and label each speaker.”
- “Please translate this audio into English.”
- “Please list what decisions were made at this meeting.”
Prompts can include product terminology and formatting requirements, helping the model handle specialized vocabulary in fields such as healthcare, financial services and customer service. Because the model’s weights are open, developers can choose where to process audio, from a desktop computer to the cloud.
According to the Hugging Face open ASR leaderboard, Nemotron 3.5 Transcribe ranks No. 2 among open-weight models for artificial analysis. NVIDIA says it delivers high accuracy with approximately 4.3 billion active parameters—80% fewer than the No. 1 open-weight model—making local, high-quality transcription more practical on available hardware.
The model is available today through Hugging Face. A quantized version is planned for local use on NVIDIA RTX Spark and DGX Spark systems, Jetson AGX Thor devices and NVIDIA Blackwell data-center GPUs. The NVFP4 version will be deployable through NeMo-Speech.cpp or vLLM. More information is available on the NVIDIA Technology Blog.
New controls manage GPU resources between apps and agents
RTX Spark systems can run long-running AI agents alongside everyday applications. Both workloads share the system’s GPU computing resources, and the preferred balance may change depending on whether users need immediate application performance or faster background processing.
New controls in the NVIDIA app let users choose how GPU resources are shared. Foreground applications can be prioritized for maximum responsiveness, or background agents can receive more resources to complete tasks sooner.
Users can find the feature in the NVIDIA app under System > Advanced Settings > Access Control Resource Sharing. Hardware-accelerated GPU scheduling must be enabled in Windows for the feature to work.
Liquid AI’s LFM3 model family comes to RTX Spark
Liquid AI’s LFM3 family contains liquid foundation models for text, vision, audio, omni and decision-making tasks. The models share a backbone, with base and trained versions available as open weights.
Liquid AI says the models are compact and fast. Its new 1.7-billion-parameter base model outperforms the previous 2.6-billion-parameter base model, according to the company.
LFM3 models can support local workflows involving Hermes Agent, Pi and agents running on OpenClaw. Potential tasks include prioritizing inboxes, managing calendars, reading documents and application screens, transcribing and translating audio, answering voice questions about handwritten mathematics, and routing support tickets and voice requests.
Developers can run LFM3 models on RTX Spark with native support through llama.cpp and vLLM, and can fine-tune the models locally.
Phantom Blade Zero launches with DLSS 4.5
S-GAME’s Phantom Blade Zero goes on sale October 28 at 7 p.m. PT through Steam and the Epic Games Store. Players take the role of Soul, a swordsman framed for murdering his master, and travel through a martial-arts world to uncover the conspiracy.
NVIDIA worked with S-GAME Studio and S-GAME Publishing to add ray-tracing effects to the game. On RTX Spark laptops and PCs with GeForce RTX GPUs, ray tracing provides more accurate bounced lighting, detailed reflections and realistic shadows for foliage and terrain.
NVIDIA’s DLSS dynamic multi-frame generation feature can increase frame rates by up to 6x through frame generation, or adjust the multiplier during gameplay to target a selected frame rate. More information is available at GeForce.com.
RTX Remix 1.6 brings path tracing to Minecraft and classic games
NVIDIA RTX Remix is a modding platform for remastering classic games with path tracing and DLSS. Version 1.6 was released last week and adds support for Minecraft, one of the world’s largest gaming communities.
Creators and players can render Minecraft Java Edition 1.21.1 through RTX Remix, using DLSS-accelerated path tracing for more realistic lighting and materials.
The update also adds DLSS 5 with 3D-guided neural rendering, up to 6x more multi-frame generation, native RTX Spark support and agent-assisted workflows. The Portal with RTX and Half-Life 2 RTX demos support RTX Spark. They include DLSS 3D-guided neural rendering for more realistic lighting and materials, new DLSS frame-generation performance modes and options to accelerate path tracing.
RTX Remix 1.6 also adds AI-assisted modding tools. Its redesigned ComfyUI workflow allows modders to generate physically based materials inside RTX Remix without setting up an application programming interface. Batch-completed projects and presets are intended to make the workflow more accessible to less-experienced creators. NVIDIA Agent Skills also allow compatible AI agents to work directly with RTX Remix on remastering projects.
More information is available on the RTX Remix Blog.
RTX Spark laptops target developers, creators and gamers
RTX Spark laptops are available in different screen sizes, colors and configurations, including touchscreens, pen support and two-in-one designs. NVIDIA says the systems are 14 mm thick, weigh 3 pounds and offer all-day battery life.
The slim design is made possible by placing the GPU, CPU and up to 128 GB of unified memory on one power-efficient superchip. Creative applications and AI models draw from the same memory pool, allowing workloads that previously required a desktop workstation to run on a laptop, including while unplugged.
- AI developers: Run models and agents locally on NVIDIA CUDA, the same platform used for AI in the cloud, with a deployment path to more than 150 million NVIDIA RTX AI PCs and NVIDIA GPUs in the cloud.
- Creators: Edit camera-native 4:2:2 video and render 3D scenes of up to 90 GB. Adobe Photoshop and Premiere offer native RTX Spark support through beta releases.
- Gamers: Run hundreds of games with sharper visuals and smoother, more responsive gameplay using NVIDIA DLSS, ray tracing, Reflex and G-SYNC.
RTX Spark laptops are currently available from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599. Models with 128 GB of memory are available in stores. Product and purchasing information is available on the RTX Spark product page.
More NVIDIA local-AI updates
RTX Spark helps train a Microduck robot
During his keynote at GTC Berlin, NVIDIA founder and CEO Jensen Huang demonstrated how AI agents running on RTX Spark can assist with training Hugging Face’s Microduck robot. In the workflow described by NVIDIA, Hermes Agent and Nemotron 3.5 Bolt plan the obstacle course on RTX Spark, Astra builds the 3D assets, NVIDIA Isaac Lab trains the navigation policy, and RTX Spark validates it with NVIDIA Isaac Sim.
NVIDIA has published a demonstration and a playbook for the workflow.
Nemotron 3.5 Bolt models support local agents
NVIDIA Nemotron 3.5 Bolt E4B 4GB is currently available and is designed to support natural conversations and multi-step agent tasks with 4 GB of memory. The model can power local voice assistants, game characters and embedded interfaces on RTX Spark, DGX Spark, GeForce RTX GPU and NVIDIA Jetson devices.
It supports llama.cpp, vLLM, Ollama, Unsloth and LM Studio, and works with Hermes and OpenClaw. For more demanding agent workloads, NVIDIA also identifies Nemotron 3.5 Bolt 74B-A9B 24GB for running local agents on RTX Spark.
FLUX 3 Image Klein runs locally
Black Forest Labs’ FLUX 3 Image Klein is a 5.6-billion-parameter open-weight model for local image generation and editing. NVIDIA says its optimized version runs on an RTX GPU with 8 GB of VRAM, while developers can fine-tune models on RTX Spark.
The NVIDIA NeMo AutoModel recipe can be used to customize the model for specific creative workflows.
DGX Spark 64GB is available from partners
NVIDIA DGX Spark 64GB is now available for purchase from manufacturer partners, giving developers and researchers another option for building and running AI locally. Powered by the GB10 Grace Blackwell Superchip, it combines 64 GB of unified memory with DGX OS and the NVIDIA AI software stack for agent development, inference and fine-tuning.
Updates are available through NVIDIA RTX Spark on X, Instagram, TikTok and Facebook. Readers can subscribe to the NVIDIA Local AI Newsletter. NVIDIA Workstation updates are available on LinkedIn and X.
See NVIDIA’s news and software product information for additional details.
Source: blogs.nvidia.com