At IFA in Berlin, NVIDIA confirmed that the first Windows PCs built on its RTX Spark platform will begin shipping in October. Lenovo and Acer are first out of the gate with announced systems, and NVIDIA says the rest of its OEM bench is right behind them. This is the company's first full client PC platform, not a GPU launch with a CPU cameo, and it is the most consequential new chip to hit Windows in years.
What RTX Spark actually is
RTX Spark is NVIDIA's first-generation N-series chip for Windows, called the N1X. It is a proper system-on-a-chip: an Arm-based Grace CPU stitched together with a Blackwell RTX GPU and a big pool of unified LPDDR5X memory that both processors share. NVIDIA's pitch is that this is the machine that finally makes "AI PC" mean something concrete. Up to 1 petaflop of AI compute, up to 128GB of unified memory, and the ability to run large AI models locally instead of renting time in somebody else's data center.
The "unified memory" part matters more than the petaflop number. When the CPU and GPU share one memory pool, a 70-billion-parameter model that would choke a discrete GPU with 16GB of VRAM can sit comfortably in 128GB of shared memory. That is the whole bet: your PC becomes the data center for your own AI workloads. Gaming and creative work ride along on the Blackwell GPU, which is, after all, what NVIDIA knows best.
Two configurations, two very different buyers
NVIDIA published two N1X configurations, and they are aimed at different wallets.
| Specification | N1X high-end | N1X laptop config |
|---|---|---|
| CPU | 20-core Grace | 18-core Grace |
| GPU | 6,144-core Blackwell RTX | 5,120-core Blackwell RTX |
| Unified memory | 24GB to 128GB | 24GB to 32GB |
| Memory type | LPDDR5X shared | LPDDR5X shared |
| AI compute | Up to 1 petaflop | Lower (not disclosed) |
| Form factors | Laptops, compact desktops | Laptops only |
The high-end config is the halo: 20 CPU cores, the full 6,144-core Blackwell GPU, and memory configurations up to 128GB. It ships in both laptops and compact desktops. The second config trims to 18 CPU cores and 5,120 GPU cores, caps memory at 32GB, and is laptop-only at launch, with compact desktops expected later.
That second config is arguably the more important one. A 32GB ceiling and a smaller GPU is how RTX Spark gets into laptops that cost closer to $2,000 instead of workstation money. Tom's Guide floated a speculative $1,999 to $2,499 range for lower-tier systems, which is explicitly a guess, not an NVIDIA number. Nobody has announced real prices yet, and that silence is the biggest open question hanging over the launch.
The OEM lineup is unusually serious
Here is the part that made me sit up. The launch partner list is Acer, ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, with Gigabyte expected to follow. That is essentially the entire Windows OEM bench committing to a brand-new chip platform at once.
Laptop supply chains do not turn on a dime. Six major OEMs showing up on day one means NVIDIA has been sampling this silicon since early in the year at the latest. Lenovo showed two systems at IFA: the Yoga Pro 9n laptop and a 16-inch Yoga 9n 2-in-1. Acer unveiled a small-form-factor desktop. Microsoft's own entries are the Surface Laptop Ultra, with a 15-inch mini-LED display hitting 2,000 nits of peak brightness, and a Surface RTX Spark Dev Box aimed at developers.
The thin-and-light angle is worth noting. Early chassis targets are around 14mm thick and as light as 3 pounds, which means NVIDIA is not pitching a mobile workstation. It wants RTX Spark in the premium ultraportable slot, the machine you carry every day that also happens to run a 120-billion-parameter model.
Game publishers have signed on too, with EA, Embark, and Ubisoft confirmed. That matters because the Windows-on-Arm gaming story has always been the weakest link, and having publishers in the tent early is NVIDIA trying to get ahead of it.
The software story is half the product
A chip is only as good as its software, and NVIDIA knows the Windows-on-Arm transition lives or dies on compatibility. The company announced Windows Agent Framework support for running local agents, PAIR for distributed inference across machines, and inference updates to llama.cpp and vLLM, which are the open-source workhorses people actually use to run models locally.
Microsoft has done its part in Windows 11: a Workload Profile Scheduling system that spreads tasks across the chip's 20 CPU cores, and a tuned Prism emulation layer for running x86 apps on Arm. NVIDIA's claim is blunt: every Windows 11 app will run. That is a big promise, and October is when we find out whether it holds.
The honest version: Arm64 driver maturity, sustained battery and thermal performance under real workloads, and independent gaming and AI benchmarks are all still open questions. The spec sheet is exciting. The reviews will be the verdict.
The memory market could spoil the pricing
One more thing to watch, and it is not NVIDIA's fault. The memory market has gotten brutal. AI data center demand has soaked up supply, and there are credible reports of RAM now accounting for roughly 60 percent of the bill of materials on some laptops, up sharply from prior years. A platform whose headline configuration ships with up to 128GB of LPDDR5X is directly exposed to that.
If memory costs keep climbing into the fourth quarter, even the estimated $2,000-ish entry point could prove optimistic, especially on higher-memory configurations. NVIDIA has not published an MSRP for either N1X config, which gives it room to maneuver, but it also means buyers are flying blind until OEMs put up prices.
What I am watching
Three things will decide whether RTX Spark is a genuine inflection point or just a very expensive tech demo. First, real prices, because everything about this platform's fate depends on whether normal people can afford it. Second, the Windows-on-Arm compatibility story in practice, not on a slide. Third, independent benchmarks of local AI workloads against the hype, because "1 petaflop" is a marketing number until someone measures it.
The biggest PC hardware launch of the fall is NVIDIA's, not Intel's or AMD's. If local AI on Windows is going to be a real category instead of a keynote bullet point, it starts here, in October, with shipping hardware you can actually buy.
Bottom line: RTX Spark is the most ambitious new Windows chip in years, the OEM support is real, and the software story is credible. Now it has to survive contact with prices, thermals, and app compatibility.