Nvidia RTX Spark Reviews Are In: Great for Local AI, Not for Gaming

The first reviews of laptops built on Nvidia’s RTX Spark platform arrived this week, and they agree on one thing: the chip is excellent at running AI models locally, but it is not a replacement for a gaming laptop. RTX Spark, also called N1X, is Nvidia’s first complete PC platform, an Arm-based chip running Windows that was announced at Computex 2026. We covered the launch of the Surface Laptop Ultra with RTX Spark and AMD’s answer, Gorgon Halo; now there are real test results.

What’s inside RTX Spark

The full N1X chip has 20 Arm CPU cores, 10 Cortex-X925 performance cores that boost to about 4GHz and 10 Cortex-A725 efficiency cores, plus a Blackwell GPU with 6,144 CUDA cores across 48 SMs, all built on TSMC’s 3nm process, according to Wccftech. CPU and GPU share up to 128GB of LPDDR5X-9400 unified memory with about 300GB/s of bandwidth, and the chip runs at 45 to 80 watts. A cut-down version with fewer cores is used in cheaper models.

Local AI: the strongest result

Tom’s Hardware tested a top-end Microsoft Surface Laptop Ultra with the full chip and 128GB of memory using AgentPerf-local, a new Artificial Analysis benchmark that measures how long an AI agent takes to finish whole tasks. With the Qwen 3.8 27B model, a 14-inch MacBook Pro with Apple’s M5 Max took about 25% longer than the Surface, and a system with AMD’s Ryzen AI Max+ 395 needed roughly twice as long, about 40 minutes. Tom’s Hardware points out that the Surface uses a bigger 140W charger than that MacBook’s 96W, and that software maturity also plays a part, but it calls RTX Spark the fastest platform at this task today.

Gaming: possible, but not the point

Gaming is where reviewers are cautious. Most Windows games are not built for Arm, so they run through Microsoft’s Prism emulation layer, which adds overhead, and the shared 300GB/s of memory bandwidth is below even the 384GB/s that a laptop RTX 5060 has to itself, Tom’s Hardware explains. In the CPU-heavy Gears of War: E-Day, GPU use averaged only about 85% at close to 1080p with frame generation on, and the game hung once during a map transition. In practice, the reviewer had to lean on both DLSS upscaling and frame generation to get a smooth experience.

PC Gamer’s ASUS ProArt P16 ran the Metro Exodus Enhanced benchmark at 75fps at 1080p, behind a Razer Blade 14 with a 115W RTX 5070 at 84fps, but while drawing an average of 43 watts compared with 97 watts. Nvidia says the chip keeps about 99% of its performance on battery, and PC Gamer’s testing bore that out.

The price problem

RTX Spark laptops start at roughly $2,600 to $2,700 depending on the maker, and PC Gamer notes the cheapest configurations come with the cut-down chip and as little as 24GB of memory. The verdicts line up: PC Gamer says it can’t recommend the platform to the average PC gamer, and Tom’s Hardware says a laptop with a dedicated RTX 50-series GPU still offers a better gaming experience, especially for anyone who doesn’t care about local AI.

Who it’s for

RTX Spark makes the most sense for developers, creators and anyone who wants to run large AI models on a laptop instead of in the cloud, thanks to its large pool of unified memory. Battery life and full reviews of individual laptops are still to come: Tom’s Hardware says the launch embargo kept it from publishing system-specific results such as battery life and display quality.

The bottom line

The first RTX Spark laptops deliver on Nvidia’s AI pitch, with local AI results that beat Apple’s and AMD’s best laptop chips in early testing. For gaming, they work but trail cheaper laptops with dedicated graphics, and prices starting around $2,600 make them a niche choice for now.

Sources: Tom’s Hardware, PC Gamer, Wccftech