Nvidia’s RTX Spark laptops went on sale Friday, October 16, from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599, the company announced. It’s Nvidia’s biggest swing yet at the Windows laptop as a whole computer rather than just the graphics card inside it.
At the center is a single “superchip,” the N1X, that combines a Blackwell RTX GPU with an Nvidia Grace CPU, which MediaTek helped design. It offers up to 128GB of unified memory shared between CPU and GPU, and up to 1 petaflop of FP4 AI performance.
The first reviews are in, and they land roughly where you’d expect: a genuinely impressive GPU, a CPU that trails Apple, battery life that doesn’t make it through the day, and prices that make your accountant wince.
What you can buy
The launch lineup per Nvidia: 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 Prestige N16 Flip AI+ and Prestige N16. Pricing and configurations vary by region.
Two versions of the chip exist. The full N1X has 20 CPU cores and 6,144 CUDA GPU cores. A cut-down 18-core, 5,120-CUDA-core version powers the cheaper configs, The Verge reports.
What it actually costs
That $2,599 floor buys a Surface Laptop Ultra or ASUS ProArt P14 with 24GB of memory, per The Verge. The configurations reviewers tested ranged from $3,199.99 for a 32GB HP OmniBook Ultra to $6,999.99 for a 128GB ASUS ProArt P16. A 128GB Surface Laptop Ultra runs $5,899.99, according to Tom’s Hardware, and Dell’s XPS 16 Creator Edition starts at $3,799.99, per Dell.
So the headline feature, 128GB of unified memory, lives at roughly double the advertised starting price. Part of that is the market: The Verge’s reviewers point to soaring memory and storage prices driven by AI demand.
The pitch: run the AI on your lap
The selling point is memory. Most laptop GPUs top out at a fraction of 128GB, which caps the size of AI model you can run locally. Unified memory lets a much bigger model fit, and Nvidia says creators can render 3D scenes of 90GB or more.
Nvidia leaned hard on agents. It lists guided installers for GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy “without requiring a command line,” plus its own Nemotron 3.5 Bolt 74B-A9B model tuned for tool calling. Its OpenShell runtime and Microsoft Execution Containers are meant to limit what those agents can touch.
There’s also a feature called PAIR that lets an RTX Spark laptop serve AI inference to other Windows, Linux and Mac computers on a home network. Your laptop, in other words, can moonlight as a tiny server.
What reviewers found: local AI is still a beta
This is where the pitch meets the fans. The Verge’s Tom Warren said local AI “feels like a beta aimed at developers right now,” with agent installers flashing up command prompts and fans “blasting away for the most basic of queries.”
Sean Hollister asked the Hermes agent, running a 27-billion-parameter Qwen model, to build a daily news briefing like the one Nvidia demoed on stage. It took 10 minutes and 10% of a 99.9 watt-hour battery, and cloud-based Claude did it in half the time with better sourcing. A reinstall made it worse: the same prompt then ate 60% of his battery. Antonio Di Benedetto burned 30% of a Lenovo’s battery in 36 minutes getting a local model to resize five product images, which it cropped wrong on the first try.
The real case, the reviewers agree, is privacy and overnight agent work for developers who don’t want their files going to a cloud service. That’s a real niche. It’s not yet a mass-market reason to buy.
Speed: great GPU, so-so CPU
On graphics, the N1X is legit. It tied Apple’s top-end M5 Max in a Blender render test and beat it in Procyon image generation in The Verge’s benchmarks. In games, Warren found medium settings at 1440p with DLSS upscaling gave a stable 60 fps in most titles, including 66 fps in Cyberpunk 2077. Di Benedetto put the GPU somewhere between an RTX 5070 and RTX 5060 laptop chip, short of Nvidia’s own RTX 5070-class claim.
The CPU is the weak spot. In Geekbench 6 single-core, the 128GB ASUS scored 3,170, against 4,330 for a MacBook Pro with M5 Max. “Even a lowly 12-core M6 Mac Mini competes with it in multicore performance,” Di Benedetto wrote, “and a Mac Mini starts at just $900.” Some CPU-heavy games, like Helldivers 2, couldn’t hold 60 fps no matter the settings.
The Arm question
The Grace CPU is Arm-based, and Windows on Arm has a long history of “works, mostly.” Nvidia and Microsoft clearly saw that coming. They list the first PyTorch-CUDA build for Windows on Arm, a native-by-default Python build, a native Conda installer and support for 97% of the top 15,000 Python packages.
The good news from reviewers: years of Qualcomm-driven work on Windows on Arm have paid off. Hollister used an RTX Spark as his daily driver with Signal, Discord, Slack, CapCut and more, and “never once needed to think whether I was running an Arm or an x86 app.” The rough edges showed up in gaming, VR drivers and AI tooling instead.
That 97% Python figure is a real effort to fix the developer story. The other 3% will be somebody’s entire workflow.
Battery life: “all-day” needs an asterisk
Nvidia promised “all-day” battery life. Tom’s Hardware measured about 10 hours on its Surface Laptop Ultra tests. But on The Verge’s ASUS ProArt P16, with a 99.9 watt-hour battery, the best real-world workday was 5 hours 40 minutes, and video editing cut it to about 5. Run a local model and, in Di Benedetto’s words, “you won’t even make it to lunch.”
The catch
Our read: “up to” is doing a lot of work here. Up to 128GB, up to 1 petaflop, and FP4, the lowest-precision number format Nvidia quotes, which flatters the figure. The $2,599 starting price doesn’t buy the maxed-out memory, the headline AI use case is still rough, and the CPU trails Apple’s by a wide margin in single-core work.
Who it’s for
- AI developers who want privacy: the 128GB models are the most capable Windows laptops for running big local models, if you can stomach the price and the fan noise.
- Creators: strong GPU rendering, but MacBook Pros remain more well-rounded and often faster in video apps.
- Gamers: a cheaper laptop with a discrete Nvidia GPU still goes further.
The takeaway
This is the most serious attempt yet to make an Arm Windows laptop a developer and AI machine, with the full weight of CUDA behind it. It also puts Nvidia in direct competition with Intel, AMD, Qualcomm and Apple for the whole PC, not just the GPU slot. Version one proves the GPU; the CPU, price and battery need a version two.
What to watch: full battery rundown tests, cheaper configurations and rumored lower-end chips, how fast local agent software matures, and whether PC buyers in a shrinking market pay up to $7,000 for any of it.
Image: Daniel Lu (User:dllu) / Wikimedia Commons, CC BY-SA 4.0
