NVIDIA used its IFA 2026 keynote in Berlin to put a firm date on a product category it has never shipped before: a full Windows PC platform built entirely around its own silicon. On September 3, 2026, the company confirmed that RTX Spark PCs will start reaching store shelves in October, and it published the first official specifications for the N1X system-on-chip that powers them, according to Wccftech. Two configurations are coming, one built for high-end laptops and compact desktops, the other trimmed down for thinner, cheaper laptops.
The announcement matters beyond spec-sheet trivia. NVIDIA has spent three decades selling graphics chips to other people’s computers. RTX Spark is the company’s attempt to sell the whole computer, wrapped around the same Arm-based Grace CPU and Blackwell GPU architecture it already uses in its DGX Spark developer boxes. If it works, NVIDIA becomes a direct rival to Apple’s Mac lineup and AMD’s Ryzen AI laptops rather than just their component supplier.
NVIDIA Confirms RTX Spark PCs Ship in October With Two N1X Configurations
NVIDIA told press at IFA that OEM partners will begin RTX Spark shipments in October, with availability rolling out region by region rather than everywhere at once, per Wccftech’s reporting. Some manufacturers will launch in a handful of markets first before expanding globally, and NVIDIA said the exact rollout schedule will vary by partner.
The reveal follows months of leaks. Notebookcheck published leaked N1X specifications back in May, and Wccftech reported CUDA core counts pulled from Windows 11 driver files in July. What changed on September 3 is that NVIDIA itself put a name, a launch window, and confirmed core counts behind the rumors. That is a meaningful shift for a company entering a market where Apple, AMD, and Qualcomm already have entrenched Windows-on-Arm and Arm-Mac products, a shift Windows Central’s coverage of the same briefing also flagged as a first for NVIDIA.
Inside the N1X SoC: Grace CPU Meets Blackwell RTX Graphics
N1X is NVIDIA’s first client-focused system-on-chip, pairing an Arm-based Grace CPU with a Blackwell RTX GPU on a single package, plus unified LPDDR5X memory shared between both, as summarized on the platform’s Wikipedia entry tracking the rollout. That architecture is the same basic idea behind the GB10 superchip in NVIDIA’s DGX Spark developer appliance, just tuned for Windows 11 instead of the Linux-based DGX OS.
NVIDIA is positioning N1X as a three-in-one chip: gaming, content creation, and on-device AI, all handled by one piece of silicon instead of a separate CPU, discrete GPU, and NPU. That pitch depends entirely on how the two announced configurations perform once independent reviewers get hardware in hand, which won’t happen until closer to the October launch.
Configuration One: 20-Core CPU, 6,144-Core GPU, Up to 128GB Memory
The higher of the two N1X tiers pairs a 20-core Grace CPU with a 6,144-core Blackwell RTX GPU, and unified memory that scales from 24GB up to 128GB depending on the OEM’s configuration, according to Wccftech. This is the tier NVIDIA is positioning for both laptops and compact desktop PCs at launch, and it is also the closest match to the GB10 chip already shipping inside DGX Spark developer boxes.
Wccftech’s report notes that the 6,144 CUDA core count lines up with NVIDIA’s discrete RTX 5070 desktop GPU, giving buyers a rough performance anchor even before independent benchmarks arrive. VideoCardz’s spec breakdown of the same announcement confirms this top tier will be offered in both a laptop SKU and a separate desktop SKU, both carrying the 20-core CPU and 6,144-core GPU pairing.
Configuration Two: 18-Core CPU, 5,120-Core GPU, Up to 32GB Memory
The second N1X tier steps down to an 18-core Grace CPU and a 5,120-core Blackwell RTX GPU, with unified memory capped at either 24GB or 32GB. Wccftech reports this configuration is laptop-only at launch, though compact desktop designs built around it are expected to follow later. NVIDIA’s own comparison, relayed by Wccftech, places its CUDA core count above the discrete RTX 5060 Ti, which suggests OEMs are aiming this tier at thinner, more affordable machines rather than raw workstation power.
The gap between 128GB and 32GB matters more than it might look on paper. Large local AI models lean almost entirely on how much unified memory a system has, so the two N1X tiers are really aimed at two different buyers: one who wants to run big models and heavy creative workloads locally, and one who wants solid 1440p gaming and everyday productivity in a lighter chassis.
RTX Spark N1X Specifications Compared
| Spec | N1X Config 1 (High-end) | N1X Config 2 (Value) |
|---|---|---|
| CPU cores | 20-core Grace CPU | 18-core Grace CPU |
| GPU cores | 6,144-core Blackwell RTX GPU | 5,120-core Blackwell RTX GPU |
| Roughly matches discrete GPU | RTX 5070 (CUDA core count) | Above RTX 5060 Ti (CUDA core count) |
| Unified memory range | 24GB up to 128GB | 24GB or 32GB |
| Form factors at launch | Laptops and compact desktops | Laptops only (desktops planned later) |
| Operating system | Windows 11 (Arm) | Windows 11 (Arm) |
| Claimed chassis thickness | As thin as 14mm on some designs | As thin as 14mm on some designs |
| Shipping window | October 2026 | October 2026 |
The three confirmed SKUs, per VideoCardz’s breakdown of NVIDIA’s announcement, are an N1X laptop with the 6,144-core GPU, an N1X laptop with the 5,120-core GPU, and an N1X desktop that only comes in the 6,144-core configuration for now. Pricing for any of the three has not been disclosed.
Which OEMs Are Shipping RTX Spark Laptops and Mini PCs
NVIDIA confirmed a broad OEM roster at IFA: Acer, ASUS, Dell, Gigabyte, HP, MSI, Lenovo, and Microsoft, according to Wccftech. That is a wider list than NVIDIA’s DGX Spark launch partners, and it includes Microsoft itself, which is notable given the Surface line’s history with Arm-based Windows devices going back to the original Surface RT.
This site previously covered an earlier RTX Spark OEM and pricing leak naming six manufacturers ahead of official confirmation, and reported on ASUS and MSI RTX Spark laptop preorders selling out before specs were even final. The September 3 announcement is the first time NVIDIA itself has put a complete, on-record OEM list behind the product, rather than leaving it to leaks and preorder pages.
NVIDIA told Wccftech that availability will differ by manufacturer, with some OEMs launching in a smaller set of regions initially and others going global from day one. That staggered approach is typical for a first-generation Windows-on-Arm push, since driver maturity and app compatibility testing tend to vary between partners.
Pricing: What’s Confirmed and What’s Still a Guess
NVIDIA’s official September 3 announcement left pricing undisclosed for both N1X tiers, according to Wccftech, which noted that the range of memory configurations across two chip variants means the eventual price spread across OEMs will likely be wide. That is a deliberate ambiguity: with eight OEMs building their own designs around the same two chips, final retail pricing depends on each manufacturer’s chassis, display, and memory choices.
Earlier reporting on this site flagged a leaked estimate of roughly $2,899 for a high-end RTX Spark configuration, based on partner sourcing ahead of the official reveal. That figure has not been confirmed or repeated in NVIDIA’s own September 3 statement, so it should be read as a pre-launch estimate rather than a locked price. For comparison, NVIDIA’s own DGX Spark developer appliance, built around the same 20-core, 6,144-core, 128GB silicon, launched with pricing that NVIDIA later raised from $3,999 to $4,699 citing memory supply constraints. RTX Spark is a different, consumer-facing product line, but that price movement on the developer side is a reminder that memory costs are a live variable for anything built on this chip family.
From DGX Spark to RTX Spark: NVIDIA’s Strategy Shift
DGX Spark shipped first, aimed at AI developers who wanted a desk-side box for local model experimentation running NVIDIA’s DGX OS on Linux. RTX Spark takes closely related silicon and repositions it for a consumer audience running Windows 11, with gaming and content creation front and center alongside AI. NVIDIA’s own framing, relayed in coverage of its Computex materials and echoed on the company’s official blog, describes RTX Spark as built on the same underlying system as DGX Spark but optimized for a different platform in both hardware tuning and software stack.
That reuse of a single silicon architecture across a developer appliance and a consumer PC platform is a departure from how NVIDIA has historically operated. The company has typically kept its data-center and workstation silicon (Grace Hopper, Grace Blackwell) separate from anything sold in a retail laptop. N1X blurs that line, and it’s a sign NVIDIA sees the same Arm CPU plus Blackwell GPU formula as flexible enough to cover both a $4,699 developer workstation and a thin-and-light Windows laptop.
Gaming Claims: 100 FPS at 1440p, DLSS 5, and Multi-Frame Generation
NVIDIA is pitching RTX Spark as the first fully capable gaming platform built on Windows-on-Arm, according to Wccftech’s coverage of the announcement. The company claims RTX Spark laptops will deliver around 100 FPS at 1440p with ray tracing and DLSS enabled, backed by DLSS 5, DLSS 4.5 Ray Reconstruction, and multi-frame generation support up to 6x. NVIDIA has also lined up game developer and publisher partnerships for the platform that it discussed at Gamescom ahead of the IFA hardware reveal.
Windows-on-Arm has struggled with game compatibility for years, largely because titles built for x86 need emulation or native Arm ports to run well. NVIDIA pairing its own upscaling and frame-generation stack directly with the N1X GPU is one way to paper over raw horsepower gaps, but it does not solve the underlying compatibility question. Reviewers will need actual hardware to test how many existing PC games run natively versus through emulation once RTX Spark units ship in October.
The Local AI Pitch: Big Unified Memory for Big Models
NVIDIA says RTX Spark’s full AI stack can run local agents with up to 120 billion parameters and context lengths up to 1 million tokens, per Wccftech’s report, supported by the Windows Agent Framework and the platform’s unified memory design. On the content creation side, NVIDIA claims RTX Spark can render 90GB 3D scenes using OptiX and ray tracing, with hardware-accelerated 4:2:2 video encoding for creators.
This is where the 128GB memory ceiling on the top N1X configuration becomes the real story, more than the CUDA core count. Local large language model inference is bottlenecked by how much memory a model and its context window can fit into, not just raw compute. This site’s earlier coverage of local AI performance gains on 24GB-and-up RTX GPUs found that memory headroom, more than clock speed, determines which model sizes are even usable on a given machine. A 128GB unified pool changes that math considerably, putting far larger open models within reach of a single machine without a multi-GPU workstation.
RTX Spark vs Apple Silicon vs AMD Strix Halo
RTX Spark is not entering an empty field. Apple’s Mac Studio, powered by the M5 Ultra chip this site covered at launch, tops out at an 80-core GPU and 512GB of unified memory, aimed at the same “big local model, one machine” audience. AMD’s Ryzen AI Max+ 395 (Strix Halo) pairs a 16-core Zen 5 CPU with a 40-compute-unit Radeon 8060S integrated GPU and up to 128GB of LPDDR5X-8000 unified memory across a 256-bit bus, already shipping in mini PCs from several vendors.
| Platform | Maker | CPU / SoC | Unified Memory Ceiling | Primary OS |
|---|---|---|---|---|
| RTX Spark N1X (top tier) | NVIDIA (via OEMs) | 20-core Grace CPU + 6,144-core Blackwell GPU | 128GB | Windows 11 (Arm) |
| RTX Spark N1X (value tier) | NVIDIA (via OEMs) | 18-core Grace CPU + 5,120-core Blackwell GPU | 32GB | Windows 11 (Arm) |
| DGX Spark | NVIDIA | 20-core Grace CPU + 6,144-core Blackwell GPU | 128GB | DGX OS (Linux) |
| Mac Studio (M5 Ultra) | Apple | 80-core GPU M5 Ultra | 512GB | macOS |
| Strix Halo mini PCs | AMD (via OEMs) | Ryzen AI Max+ 395, 16-core Zen 5 CPU + 40-CU Radeon 8060S | 128GB | Windows 11 / Linux |
The comparison isn’t a clean win for anyone. Apple’s memory ceiling is far higher, but macOS runs a narrower game library than Windows. AMD’s Strix Halo already ships today with a comparable memory ceiling to RTX Spark’s top tier, and it runs standard x86 Windows rather than an Arm build that depends on emulation for legacy software. NVIDIA’s advantage is DLSS and its broader AI software ecosystem, built up over more than a decade of CUDA dominance, which neither Apple nor AMD can fully match on the software side.
Market Impact: Why NVIDIA Wants a Piece of the Windows PC Business
NVIDIA just posted a record quarter, reported at $96.2 billion as this site covered in an earlier report on Nvidia’s earnings, driven overwhelmingly by data-center GPU sales rather than consumer hardware. Entering the client PC market as a full-system vendor, rather than a component supplier selling GPUs to Dell or ASUS, gives NVIDIA a second consumer-facing revenue line that doesn’t depend on hyperscaler AI infrastructure budgets holding up. It also puts NVIDIA in more direct competition with the same OEMs it currently sells chips to, since those same eight companies are now building machines competing with NVIDIA’s own AI software ecosystem for developer mindshare.
For the eight confirmed OEM partners, RTX Spark is a chance to sell a Windows-on-Arm machine with a brand name (NVIDIA, RTX, GeForce) that already carries weight with gamers and creators, something Qualcomm’s Snapdragon X series has struggled to build despite a two-year head start in the Windows-on-Arm laptop space.
Sizing a Local Model to Unified Memory
For readers evaluating whether an RTX Spark configuration, a Strix Halo mini PC, or a Mac Studio fits a local AI workload, the deciding factor is almost always memory headroom rather than raw core counts. A rough rule used by the local inference community is that a quantized model needs roughly its parameter count in gigabytes at 8-bit precision, or about half that at 4-bit, plus overhead for context length.
# Rough local-inference memory budget check
# model_params_billion * bytes_per_param + context_overhead_gb
MODEL_PARAMS_B=70 # e.g. a 70B parameter open model
QUANT_BYTES=0.5 # ~4-bit quantization
CONTEXT_OVERHEAD_GB=6 # long-context KV cache estimate
python3 -c "
params = $MODEL_PARAMS_B
bytes_per_param = $QUANT_BYTES
overhead = $CONTEXT_OVERHEAD_GB
required_gb = params * bytes_per_param + overhead
print(f'Approx unified memory needed: {required_gb:.1f} GB')
"Run that math against a 70-billion-parameter model at 4-bit quantization, and the requirement lands around 41GB, comfortably inside the top N1X tier’s 128GB ceiling but well past the 32GB value tier or a 24GB entry configuration. That gap is exactly why NVIDIA, Apple, and AMD are all racing to push unified memory ceilings higher rather than just adding CPU or GPU cores, a trend worth tracking alongside the rest of the hardware news this site covers week to week.
What Comes Next: Five Predictions for RTX Spark
Based on NVIDIA’s confirmed roadmap and the competitive landscape it’s stepping into, here’s how the next two quarters likely play out.
- Pricing for the top 128GB N1X configuration will land above $2,000 once OEMs disclose SKUs closer to October, given the memory-driven price hikes NVIDIA already absorbed on DGX Spark.
- Game compatibility, not raw performance, will be the dominant early review theme, since Windows-on-Arm emulation has been the recurring weak point for every prior Arm-based Windows laptop.
- At least one of the eight confirmed OEMs will delay its regional launch beyond October, consistent with NVIDIA’s own acknowledgment that availability will vary by partner.
- AMD and Apple will use RTX Spark’s launch as a marketing opportunity to highlight their own unified-memory ceilings, particularly Apple’s 512GB Mac Studio option, which dwarfs both N1X tiers.
- NVIDIA will expand the desktop-only N1X 6,144-core SKU into a broader mini PC lineup by early 2027, following the same trajectory DGX Spark took from developer appliance to wider availability.
Frequently Asked Questions
When do RTX Spark PCs go on sale?
NVIDIA confirmed shipments begin in October 2026, though exact dates and regional availability will vary by OEM, according to Wccftech’s report on the September 3 announcement.
What is the N1X chip?
N1X is NVIDIA’s first client system-on-chip, combining an Arm-based Grace CPU with a Blackwell RTX GPU and unified LPDDR5X memory on one package, built for Windows 11 laptops and compact desktops.
What’s the difference between the two N1X configurations?
The high-end config pairs a 20-core CPU with a 6,144-core GPU and up to 128GB of memory, available in laptops and desktops. The value config pairs an 18-core CPU with a 5,120-core GPU and up to 32GB of memory, laptop-only at launch.
How much will RTX Spark PCs cost?
NVIDIA has not disclosed official pricing for either configuration as of the September 3 announcement. Earlier unofficial estimates put a high-end configuration near $2,899, but that figure predates NVIDIA’s own confirmation and should be treated as unverified until OEMs publish retail pricing.
Which companies are making RTX Spark PCs?
NVIDIA named Acer, ASUS, Dell, Gigabyte, HP, MSI, Lenovo, and Microsoft as launch OEM partners.
Is RTX Spark the same as DGX Spark?
No. Both use closely related Grace-plus-Blackwell silicon, but DGX Spark is a Linux-based developer appliance that launched earlier, while RTX Spark is a Windows 11 consumer platform aimed at gaming, content creation, and local AI.
Can RTX Spark PCs run existing PC games?
NVIDIA claims strong 1440p performance with DLSS 5 and multi-frame generation, but RTX Spark runs Windows-on-Arm, meaning older x86-only games may depend on emulation rather than native support. Independent testing after the October launch will clarify real-world compatibility.
How does RTX Spark compare to a Mac Studio for local AI?
Apple’s Mac Studio with the M5 Ultra chip supports up to 512GB of unified memory, well above RTX Spark’s 128GB ceiling on its top configuration, though NVIDIA’s CUDA software ecosystem remains more widely supported across AI development tools.
