Nvidia's RTX Spark laptops are on sale, and early tests split on the CPU

Nvidia’s first Windows laptop chip is now in shops. Laptops built on RTX Spark went on sale worldwide on October 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI, Nvidia announced, with prices starting at $2,599 in the US. A follow-up Nvidia blog post on October 28 said versions with 128GB of memory are now on shelves.

The launch list covers nine machines: the ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI’s Prestige N16 and Prestige N16 Flip AI+. Nvidia says the thinnest is 14mm and the lightest weighs 3 pounds.

What is inside the N1X

RTX Spark is the platform name. The chip in the laptops is the N1X, which puts an Arm-based Nvidia Grace CPU and a Blackwell RTX GPU on one package with shared memory. Nvidia’s press release says MediaTek worked on the custom CPU design.

The specification page lists two laptop versions. The top one has a 20-core CPU, a 6,144-core GPU, up to 128GB of LPDDR5X and 1 petaflop of FP4 AI throughput. A cut-down version has 18 CPU cores, 5,120 GPU cores, up to 32GB and 827 teraflops. Both run at 45 to 80 watts and use a 256-bit memory bus. A 140-watt desktop version is listed, but no desktops shipped at launch.

The memory is the main point. A laptop with a separate graphics card is limited to that card’s own video memory. On the N1X, the GPU can draw on a large share of a 128GB pool, an approach Apple also uses with unified memory in its M-series Macs. That lets one laptop load AI models that a discrete laptop GPU cannot hold.

Nvidia frames the machines as tools for local AI, creative work and games. Founder and CEO Jensen Huang made the pitch in the launch release:

“RTX Spark brings the power of NVIDIA AI, RTX graphics and CUDA together with Microsoft Windows in one PC” (NVIDIA Newsroom)

On software, Nvidia says it shipped the first PyTorch-CUDA build for Windows on Arm and claims support for 97% of the top 15,000 Python packages. For games, it promises AAA titles at 1440p and over 100 frames per second, with Cyberpunk 2077, Fortnite and World of Warcraft among launch titles.

What early testing shows

Two detailed tests of the Surface Laptop Ultra reach different conclusions about the CPU, and the reason is the type of work measured.

PCWorld’s Mark Hachman focused on everyday use and found single-core speed weak, with settings menus that felt slow. His summary of the platform:

“a surprisingly average CPU, paired with a terrific GPU and good battery life” (PCWorld)

He measured about 10 hours of battery life in an office-work benchmark. In ComfyUI’s built-in benchmark, the laptop landed above an RTX 5060 laptop and closer to an RTX 5090 machine. Adobe Premiere only worked in a beta build.

StorageReview ran heavier workloads and found the opposite on multithreaded work. The N1X scored 6,963 in Cinebench 2026 multi-core, about 34% ahead of a Razer Blade 16 and 47% ahead of a Dell Pro Precision 5 16S. It compiled LLVM in 206 seconds against 327 for the Blade. Single-thread scores still trailed the x86 laptops, and memory latency was the worst in the group.

The local AI results show both the strength and the limit. With 64GB assigned to the GPU (62.5GB usable), StorageReview ran 20B and 35B-parameter models at roughly 58 to 89 tokens per second. On a small 8B model that fits in its own memory, the Blade’s RTX 5090 generated text about 2.3 times faster. The N1X wins on capacity, not raw speed.

StorageReview also hit a serious fault. An all-core math solver cut power to the laptop three times, and two review units failed the same way. Logs pointed to a hard power loss rather than overheating, and the cause was unresolved at publication.

Who should care

The N1X makes most sense for developers and researchers who want to run mid-sized open models on a laptop without a cloud bill, and for 3D or video work that needs more GPU memory than a mobile graphics card offers. For that buyer, the closest rival is a high-memory MacBook Pro, and the N1X adds CUDA, the Nvidia software stack that PyTorch builds and tools such as ComfyUI target.

It makes less sense for anyone whose day is browsers, office apps and light editing. Those tasks lean on single-core speed, where both tests found the chip behind. At $2,599 and up, a cheaper x86 or Apple laptop is likely to feel quicker for that work.

Several things are still unknown. Both tests cited here used Microsoft’s Surface Laptop Ultra, and other designs may run cooler or louder. StorageReview’s shutdown problem needs an explanation from Microsoft or Nvidia. Neither test checked anti-cheat software, which multiplayer gamers on Windows on Arm will want answered. Nvidia has also not given a date for desktop systems.