NVIDIA RTX Spark laptops launch with up to 128GB of unified memory

What happened

NVIDIA launched RTX Spark laptops on October 16, 2026, bringing its Grace CPU and Blackwell RTX GPU technology together in Windows PCs. The company names ASUS, Dell, HP, Lenovo, Microsoft, and MSI as launch partners and says additional systems will follow.

NVIDIA says it held launch celebrations in San Francisco, Berlin, Taipei, Beijing, Istanbul, and Seoul. In Berlin, Jensen Huang made a surprise appearance and gave away five personally signed RTX Spark laptops.

PCMag reported a line of more than 100 people before the San Francisco event, but observed only a handful of purchases. Several attendees cited the laptops’ cost as a reason not to buy that day.

NVIDIA positions the platform around running AI models and agents on the laptop itself. It says guided setup and installers support agents including GitHub Copilot, Hermes, and OpenClaw without requiring users to work through a command line.

NVIDIA says RTX Spark supports local coding, testing, and prototyping through native CUDA and Windows Subsystem for Linux. Its Windows Developer Config is intended to simplify setup, while the NVIDIA Skills repository teaches compatible agents to use CUDA-X tools and libraries.

The company also offers Nemotron 3.5 Bolt 74B-A9B 24GB, a model intended for local agents. NVIDIA says it uses 2-bit quantization to fit a 74-billion-parameter mixture-of-experts model into 24GB and is available through tools including llama.cpp, Ollama, and LM Studio.

For agent permissions, NVIDIA points to its OpenShell runtime and Microsoft Execution Containers. It also says NVIDIA PAIR allows an RTX Spark system to serve AI inference to compatible Windows, Linux, and macOS computers on a home network.

Key specs

ThePCEnthusiast reports that the 18-core N1X supports up to 32GB of unified memory, while the 20-core version supports up to 128GB. The platform’s maximum memory specification therefore does not apply to both processor versions.

A 32GB laptop does not necessarily use the smaller processor: ThePCEnthusiast lists MSI’s base Prestige N16 Flip AI+ with the full 20-core CPU and 6,144-core GPU. Its base configuration includes a 1TB SSD, a 2880 x 1800 OLED touchscreen with refresh rates up to 120Hz, a 360-degree hinge, and a stylus stored and charged inside the laptop.

Pricing and availability

NVIDIA’s October 17 launch recap lists RTX Spark laptops as starting at $2,399, without naming the configuration at that price. That differs from the $2,599 entry price reported by Tom’s Hardware at launch.

Tom’s Hardware reports that Best Buy’s launch offerings include a 24GB ASUS ProArt P14 at $2,599 and a 48GB Surface Laptop Ultra at $3,999. Higher-memory configurations are also available directly from manufacturers.

The same report lists the ASUS ProArt P16 at $4,499.99 with 64GB of unified memory, rising to $6,999.99 for a 128GB configuration with 2TB of storage. It lists HP’s 128GB OmniBook X at $5,109 and OmniBook Ultra at $5,309, both with 1TB of storage.

Buyers also need to check processor specifications rather than relying on the RTX Spark N1X name alone. Tom’s Hardware reports a full configuration with a 20-core CPU and 6,144-core GPU, alongside a smaller configuration with an 18-core CPU and 5,120-core GPU.

MSI availability is more specific than NVIDIA’s launch announcement suggests. While NVIDIA named MSI among the manufacturers available at launch, ThePCEnthusiast reports that the US Prestige N16 Flip AI+ base model is listed for preorder at $3,299 with a November 6, 2026, release date; shipping and delivery dates may vary.

ThePCEnthusiast also lists MSI configurations at $4,199.99 for 64GB of memory and a 1TB SSD, and $6,799.99 for 128GB and a 2TB SSD. The flagship version includes a 3840 x 2400 Tandem OLED touchscreen rather than the base model’s 2.8K OLED panel.

Why it matters

• Local development

  NVIDIA says native CUDA support lets developers run models and coding agents locally while keeping locally processed code and context on their device.

• Larger workloads

  NVIDIA says the shared memory supports demanding creative projects and local AI models running simultaneously.

• Software support

  NVIDIA says its work with Microsoft adds a PyTorch-CUDA build for Windows on Arm and Arm-compatible versions of more than 97% of the top 15,000 Python packages.

Early testing

The Verge tested five laptop configurations from Microsoft, ASUS, HP, and Lenovo over a week. Its writers generally found everyday Windows applications usable, including through Microsoft’s Prism emulation layer, although they reported occasional loading problems and application quirks.

Gaming results were less consistent. The Verge’s Tom Warren reported that medium settings at 1440p with DLSS performance mode generally produced a stable 60fps in most games he tested. Sean Hollister reported CPU-related limits in some titles, including battle frame rates between 37fps and 39fps in Star Wars Zero Company regardless of graphics settings.

Wccftech’s initial ASUS ProArt P16 testing found that most of the local language models it tried ran adequately, and it reported leading performance in its UL Procyon AI comparison. Its broader benchmark results varied by workload, with RTX Spark trailing some competing chips in office and Blender tests.

Tom’s Hardware tested a Surface Laptop Ultra with 128GB of memory using AgentPerf-local, which replays agent inference tasks and measures completion time. Its 14-inch M5 Max MacBook Pro took about 25% longer than the Surface on the Qwen 3.8 27B workload; on Qwen 3.6 35B A3B, the Surface finished in eight minutes versus the MacBook Pro’s ten.

Those tests used llama.cpp builds with CUDA, Metal, or ROCm acceleration as appropriate. Tom’s Hardware also noted that the Surface was a 15-inch system with a 140W adapter, compared with the tested MacBook Pro’s 14-inch chassis and 96W adapter; the results do not establish performance across every laptop configuration or inference engine.

Notebookcheck’s review of the 128GB ASUS ProArt P16 found that it retained full performance on battery, but criticized its battery runtime. The review also reported a 4K Tandem OLED touchscreen running at 120Hz, two SSD slots, and a 1080p webcam that fell short of its expectations for the price.

Who it is for

• AI developers who want to run coding agents, test applications, and customize models locally.
• Creators working with video editing, image editing, and large 3D scenes.
• Gamers interested in portable Windows systems with NVIDIA RTX graphics and DLSS support.