RTX Spark Laptops Compared: Brilliant at AI, Beaten at Games

On October 16, Nvidia stopped being a company that sells parts for other people’s laptops and became a platform. RTX Spark machines from ASUS, Dell, HP, Lenovo, Microsoft, and MSI went on sale the same morning, all built on one chip: the N1X, an Arm based Grace CPU fused with a Blackwell RTX GPU and a pool of unified memory the two share. Nine models make up the first wave, and the review embargoes have now lifted. This is the comparison that matters: one chip, six badges, and a very clear split between what these machines are for and what they are not.

The chip in sixty seconds

The full N1X carries 20 Arm CPU cores (ten Cortex-X925 performance cores boosting to about 4 GHz, ten Cortex-A725 efficiency cores) and a Blackwell GPU with 6,144 CUDA cores, built on TSMC’s 3nm process and running at 45 to 80 watts. CPU and GPU share up to 128GB of LPDDR5X memory at roughly 300 GB/s of bandwidth, and Nvidia rates the package at a petaflop of FP4 AI compute. A cut down tier sits below it in cheaper models: 18 CPU cores, 5,120 GPU cores, up to 32GB of memory, 827 teraflops. That memory pool is the entire pitch. A conventional gaming laptop chokes on a large model because its GPU has 8 or 12GB of VRAM. A top tier Spark laptop can simply load the model, up to about 120 billion parameters at 4 bit, into the shared pool and run it.

The lineup and the money

The Surface Laptop Ultra is the reference machine: 15 inch mini LED touchscreen at 120Hz, the most ports ever fitted to a Surface Laptop, a user replaceable SSD, and a range that runs from the $2,599 base (18 cores, 24GB, 512GB) to $5,899 fully loaded. The ASUS ProArt P14 and P16 aim squarely at creators; reviewers liked the hardware and flagged missing Copilot+ features at launch. HP splits its bet: the OmniBook X 14 starts at $2,999 (shipping November 4) while the OmniBook Ultra was the value standout among tested configs at $3,199.99. The Lenovo Yoga 9n brings the 2-in-1 hinge, and the Dell XPS 16 starts highest at $3,799. MSI’s Prestige N16 pair rounds out the nine. The uncomfortable truth of the chart: the $2,599 entry machines ship with 24GB of memory, which makes them a different product from the 128GB flagships wearing the same badge. On this platform, the memory tier is the machine.

Local AI: the whole point

Tom’s Hardware ran the new AgentPerf-local benchmark from Artificial Analysis, which times complete AI agent tasks using the Qwen 3.8 27B model, the same model family that tops our uncensored LLMs ranking. The result is the headline of this launch.

Around that headline, hands on testing found Nvidia’s Nemotron 3.5 Bolt running at 56 to 60 tokens per second on the Surface, the mixture of experts Gemma 4 26B at around 40, and dense 27B class models needing 4 bit quantization to stay quick. Nvidia’s own launch claims (up to 2.1x faster time to first token and 6.2x faster AI video generation than an M5 Pro MacBook Pro) are vendor numbers, but the independent results point the same direction. For local inference, this is currently the laptop platform to beat.

Gaming: manage expectations

Playable, with an asterisk the size of the chart. Frame generation inflates those grey bars while input latency stays high: Tom’s Hardware measured around 84 milliseconds of PC latency in Gears of War: E-Day at a respectable 78 fps, called out at least one major hang, and saw ray tracing make both frame rate and latency worse in Control Resonant. Three structural reasons recur in every review: the shared memory’s 300 GB/s is modest next to a discrete card’s dedicated bandwidth, the Arm CPU trails Apple’s silicon on single core work, and most Windows games arrive through Microsoft’s Prism emulation layer rather than running natively. Prism itself earned real praise this cycle, and everyday apps, including native Photoshop and Premiere builds, mostly just work now. But anti-cheat support is still rolling out, and no reviewer recommends a Spark laptop as a gaming first purchase. A cheaper machine with a discrete RTX 50 series GPU remains the better games console.

The catches

Battery life is the quiet scandal of the category. The best real world result in a week of mixed testing was five hours and forty minutes of ordinary work, from machines carrying some of the largest batteries legally allowed on a plane. Reviewers also logged a 20 to 25 percent performance drop when running on battery, plus heat and fan noise under sustained AI loads that sit oddly in laptops this thin. Add launch software gaps, Copilot+ features still landing, and the memory tier trap from the pricing chart, and the first wave tax is real even by early adopter standards.

Verdict

Buy an RTX Spark laptop if your work is local AI or heavy creation and your budget reaches the high memory tiers: the platform is genuinely first at the thing it was built for, and the Surface Laptop Ultra is its best expression, with the HP OmniBook Ultra the tested value pick. Skip it, for now, if you game first or live on battery. The chip is a landmark. The laptops around it are a strong first draft, priced like a final one.

Comparison draws on launch week reviews and testing from Tom’s Hardware, The Verge, Pocket-lint, Windows Central, StorageReview, and The New Stack, plus manufacturer launch pricing, checked October 2026.