NVIDIA expands RTX Spark laptop lineup with 128GB unified memory

NVIDIA expanded the RTX Spark laptop lineup on 28 October 2026, introducing new models from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. These systems feature an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU combined into a single superchip—an integrated CPU/GPU chip—providing up to 128GB of unified memory (shared memory) and up to 1 petaflop of FP4 AI performance (low-precision AI speed). This hardware enables high-end AI development and 3D rendering to occur locally on a thin-and-light laptop rather than requiring a desktop workstation or cloud infrastructure.

If you currently rely on a laptop with a discrete GPU and separate system RAM, none of this changes your next purchase until you need to run models that exceed your current VRAM limits.

The Blackwell and Grace superchip architecture

NVIDIA uses a single superchip to house the Blackwell RTX GPU and Grace CPU, which creates a unified hardware architecture that removes the traditional memory bottleneck between the CPU and GPU. This design allows creative projects and AI models to draw from a single 128GB unified memory pool, enabling the laptop to handle tasks that previously required a desktop workstation. Because the memory is shared, the system can move data between the processor and the graphics unit without the latency of a traditional external bus.

The architecture also supports 1 petaflop of FP4 AI performance.

The architecture also supports 1 petaflop of FP4 AI performance. This high low-precision AI speed allows high-quality, intelligent transcription and AI tasks to run locally on the hardware without needing cloud scaling. By combining the Blackwell and Grace silicon, the device functions as a personal AI supercomputer in a portable form factor.

Local AI development with Nemotron 3.5 Transcribe

Developers can now deploy the Nemotron 3.5 Transcribe model—an open weight (publicly available model) speech model—directly on RTX Spark hardware. The model handles transcription, speaker attribution, and translation across 40 languages using a single promptable interface. Because the model has approximately 4.3 billion active parameters, it is 80% smaller than the leading open weight alternative, making high-quality transcription more practical for local deployment.

Developers can use vLLM to run Liquid AI's LFM3 models on the RTX Spark

Developers can use vLLM to run Liquid AI’s LFM3 models on the RTX Spark platform with native support. To begin local deployment, developers should check Hugging Face for the Nemotron 3.5 Transcribe weights. The hardware allows for local fine-tuning of these models, providing a path to deploy across more than 150 million NVIDIA RTX AI PCs and cloud GPUs using the same CUDA platform used in the cloud.

Resource management for foreground apps and agents

The NVIDIA App includes new resource management controls that allow users to balance GPU power between foreground applications and background agents. This feature allows long-running agents to work alongside active software without compromising the performance of the primary task. Users can prioritize the foreground app for maximum frame rates or allocate more resources to a background agent to complete its work faster.

To use this resource sharing, users must enable hardware-accelerated GPU scheduling in Windows Settings. This allows for a smoother experience when running agentic AI tasks, such as using LFM3 models in Hermes Agent, Pi, or OpenClaw to manage calendars and triage inboxes while the user remains active in other applications.

Workflows for 3D creators and video editors

3D artists can render 3D scenes of up to 90GB on the RTX Spark laptop, a task that typically requires a desktop workstation. The unified memory pool helps ensure that large scenes fit within the available 128GB capacity. Additionally, the system supports RTX Remix 1.6, which allows for the remastering of classic games like Minecraft with DLSS-accelerated path tracing (realistic lighting).

Video editors can process camera native 4:2:2 video on the device, as the superchip provides native support for this format. Adobe Photoshop and Premiere offer native RTX Spark support specifically through beta releases. This enables high-resolution editing in a 14mm thin and 3 lbs weight chassis, providing workstation-grade memory capacity in a highly portable form factor.

Pricing, design, and availability

NVIDIA says RTX Spark laptops are available now from ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with configurations starting at $2,599. The laptops are designed for portability, arriving in various designs including touchscreens and two-in-one convertibles. The 14mm thinness is made possible by the power-efficient superchip design, which integrates the GPU, CPU, and memory into a single component.

The 128GB configurations

The 128GB configurations are available on shelves now. While the starting price is $2,599, the availability of these high-memory models allows users to avoid the need for a separate desktop workstation for specific AI and 3D workflows. Availability is confirmed by NVIDIA, but the source does not specify the specific regions where these models are currently in stock.

Who should buy an RTX Spark laptop

AI developers who need to run models and agents locally on the same CUDA platform used in the cloud should consider the RTX Spark for its 128GB unified memory. The large memory pool allows for much larger local model deployment than standard laptop GPUs. To start, developers should verify that their local CUDA environment matches their cloud deployment environment.

Freelance video editors

Freelance video editors who need to edit camera native 4:2:2 video on a portable machine should choose the RTX Spark. The superchip supports native 4:2:2 editing and 90GB scene rendering. Editors should time their move to the new hardware for when the beta release of Premiere reaches a stable state.

3D artists who need to render large 3D scenes (up to 90GB) without a desktop workstation should use the RTX Spark. The unified memory pool allows for high-capacity rendering in a 14mm chassis. Artists should monitor their 3D scene size to ensure it stays within the 90GB limit of the RTX Spark.

Users on a budget or cloud-only workflows can skip this

Budget-conscious students should skip this product because the starting price is $2,599. For users looking for a more affordable entry point, standard RTX laptop configurations are a more suitable choice. Users who only perform cloud-based AI work should also ignore this hardware, as the system is specifically designed for local inference and local fine-tuning.

Cloud-only users should instead consult cloud GPU instance documentation to find solutions that scale beyond local hardware limits. This laptop is a specialized tool for local processing, and those who do not need to run models locally will find the cost does not offer a relevant value proposition.

A 3D artist’s workflow walkthrough

Priya is a freelance 3D artist who needs to render complex scenes for clients but travels frequently to different job sites. She purchases an RTX Spark laptop to replace her bulky desktop workstation. She uses the 128GB of unified memory to load 90GB 3D scenes that would normally require a desktop, and she uses the superchip’s power to render the scene while the laptop remains thin enough to fit in her bag. Priya can complete her work on-site and deliver high-quality renders without needing a bulky desktop workstation.

If Priya needed to render a scene larger than the 90GB limit supported by the laptop, she would have to return to a desktop workstation. Because the 90GB limit is a hardware constraint of the Spark platform, exceeding that figure would invalidate the portability benefit.

Practical applications for the RTX Spark

What the RTX Spark replaces

The RTX Spark replaces desktop workstations for 3D and AI work by providing the same memory pool and compute power in a 3lb, 14mm form factor. For developers, it also reduces the need for multiple models for transcription, translation, and speaker labeling by using a single promptable model like Nemotron 3.5 Transcribe. Finally, it replaces cloud-based transcription services, as it enables high-quality, intelligent transcription to be performed locally on the device.

The system also sits alongside the NVIDIA DGX Spark 64GB

The system also sits alongside the NVIDIA DGX Spark 64GB, which offers a smaller 64GB unified memory capacity for local AI development. While the DGX Spark is a dedicated unit, the RTX Spark laptops offer the same superchip architecture in a consumer-ready, portable chassis. The primary limitation of the Spark laptops remains the 90GB scene limit and the $2,599 starting price, which are not yet established as being lower in future iterations.

This analysis is based on NVIDIA’s NVIDIA RTX Spark Windows Laptops Gain New Models, Agent Tools and Games.