NVIDIA's first Windows laptop chip went on sale on October 16, 2026: RTX Spark laptops pair an Arm CPU with a Blackwell RTX GPU and up to 128 GB of shared memory, starting at $2,599. Nine models from ASUS, Dell, HP, Lenovo, Microsoft and MSI are in stores now. The big questions are practical: will your apps and games run on an Arm chip, what can 128 GB of memory do for local AI, and who should actually buy one? This guide answers them from NVIDIA's and Microsoft's own documents, as of October 17, 2026.
What is RTX Spark?
RTX Spark (the chip is also called N1X) is a single package with a CPU and a GPU that share one pool of memory. NVIDIA calls it a superchip. According to NVIDIA's specifications page and its Windows on Arm porting guide:
| Spec | Top laptop version | Lower laptop version |
|---|---|---|
| CPU | 20-core Grace (10 Cortex-X925 performance + 10 Cortex-A725 efficiency cores) | 18-core Grace |
| GPU | Blackwell RTX, 6,144 CUDA cores | Blackwell RTX, 5,120 CUDA cores |
| Memory | Up to 128 GB LPDDR5X, shared by CPU and GPU | Smaller maximum (NVIDIA's pages list 32 GB and 64 GB) |
| Memory bandwidth | 300 GB/s over a 256-bit bus | Not separately listed |
| AI performance (FP4) | 1 petaflop | 827 teraflops |
| Laptop power | 45 to 80 W | 45 to 80 W |
Two things make it different from a normal gaming laptop. First, the CPU uses the Arm instruction set, like a phone or an Apple silicon Mac, instead of the x86 chips from Intel and AMD that Windows PCs have used for decades. Second, there is no separate graphics card with its own small pool of video memory. The GPU reads the same memory as the CPU, which is why a laptop can offer the GPU 128 GB instead of the 8 to 24 GB that laptop graphics cards usually have. Our companion guide, What is unified memory?, explains why that matters.
Not sure what is in your current computer? What GPU Do I Have? reads the graphics chip your browser is using in a second, with no download.
Which RTX Spark laptops can you buy?
NVIDIA's launch announcement lists these models, sold on manufacturers' websites and at major retailers worldwide:
- ASUS: ProArt P14 and ProArt P16
- Dell: XPS 16
- HP: OmniBook Ultra and OmniBook X 14
- Lenovo: Yoga Pro 9n
- Microsoft: Surface Laptop Ultra
- MSI: Prestige N16 Flip AI+ and Prestige N16
NVIDIA says prices start at $2,599 and vary by region and configuration. Microsoft also announced a Surface RTX Spark Dev Box desktop for developers at $5,999, available for pre-order in the US and shipping in November.
Will my Windows apps run on RTX Spark?
Most will, but some run through translation. Windows on Arm runs three kinds of software:
- Native Arm64 apps run at full speed. Many popular apps already ship Arm versions; NVIDIA's ecosystem update lists the apps it says are ready.
- x86 and x64 apps run through Prism, Microsoft's built-in emulator. Microsoft says Prism is optimized for RTX Spark and supports the AVX and AVX2 instruction extensions that many newer programs need. Emulated apps work without changes but can run slower than native ones.
- Drivers cannot be emulated. Hardware that needs its own driver, such as some printers, audio interfaces, VPN clients and older antivirus tools, needs an Arm64 driver from its maker.
Before you buy, check your must-have software:
- Search the vendor's site for "Windows on Arm" or "ARM64" support.
- For any device you plug in, check that an Arm64 driver exists.
- If you rely on a niche or old business app, test it on an Arm PC first or ask the vendor.
Developers get the same CUDA tools they use on other NVIDIA GPUs, according to NVIDIA's porting guide, which also documents building native Arm64 versions of existing apps.
Can RTX Spark laptops play games?
Yes, with one big exception. NVIDIA says RTX Spark launched with hundreds of games that work on day one, with DLSS, Reflex, G-SYNC and ray tracing support. Its launch list includes Counter-Strike 2, Dota 2, Fortnite, Marvel Rivals and ARC Raiders, and NVIDIA released a Game Ready Driver for launch day.
The exception is anti-cheat. Many online games refuse to start unless their anti-cheat software runs, and anti-cheat often relies on a low-level driver, which, like any driver, needs an Arm64 version. NVIDIA says it is still working with BattlEye, EA Javelin, Easy Anti-Cheat and others, and that EA Javelin support arrives in early 2027. If your main game is a competitive online shooter, check that it is on NVIDIA's list before buying.
NVIDIA does not say how many of the listed games run as native Arm builds and how many run through Prism, so expect some variation in frame rates from game to game until independent reviews arrive.
What can 128 GB of shared memory do for local AI?
This is the headline feature. A large language model must fit in the memory its GPU can use, and the GPU in RTX Spark can use most of the 128 GB. NVIDIA's product page says the platform can run models of up to 120 billion parameters, a claim it has not backed with independent benchmarks.
Here is how that works out with the standard estimate (parameters × bits per weight ÷ 8), from our LLM VRAM Calculator:
| Model | Format | Weights in memory | Fits in 128 GB? |
|---|---|---|---|
| 8B dense | Q4_K_M (4.8 bits) | about 5 GB | Easily |
| 70B dense | Q4_K_M | about 42 GB | Yes |
| 120B mixture of experts | 4.5-bit | about 68 GB | Yes, with room for context |
| 816B mixture of experts (Muse Spark Open) | 2-bit | about 265 GB | No |
Memory size decides whether a model runs; memory bandwidth decides how fast. At 300 GB/s, a dense 70B model at 4-bit reads about 42 GB per generated token, which caps it near 7 tokens per second in theory. A mixture-of-experts model that reads only a few billion parameters per token is much faster. That is why NVIDIA's own local model, Nemotron 3.5 Bolt 74B-A9B, uses a mixture-of-experts design with 9 billion active parameters and 2-bit weights that fit in 24 GB. Our guide to mixture of experts explains the trade-off, and How to run Muse Spark Open locally shows the same math for a much bigger model.
RTX Spark vs a MacBook Pro with M5 Max
The closest rival is Apple's MacBook Pro, which has used shared memory for years. Both top out at 128 GB:
| RTX Spark (top version) | M5 Max MacBook Pro | |
|---|---|---|
| Max memory | 128 GB | 128 GB |
| Memory bandwidth | 300 GB/s | Up to 614 GB/s |
| GPU software | CUDA, DirectX | Metal, MLX |
| Operating system | Windows 11 on Arm | macOS |
| Starting price | $2,599 (NVIDIA) | Varies by configuration |
Apple's figures come from its M5 Pro and M5 Max announcement. On paper the Mac has about twice the memory bandwidth, which matters most for generating text with large dense models. RTX Spark's advantages are Windows, CUDA, the GPU platform most AI tools support, and NVIDIA's game and driver support. Neither set of spec sheets tells you real-world speed; wait for independent tests if speed is the deciding factor.
Who should buy an RTX Spark laptop?
- Good fit: developers and researchers who want to run 30B to 120B models locally on Windows with CUDA; creators whose apps already have Arm versions; people who want a thin laptop that also games.
- Wait: players of competitive online games that need anti-cheat not yet supported; anyone who depends on x86-only drivers or niche business software; buyers who want independent battery-life and performance reviews first.
- Not needed: if you mainly browse, write and stream video, a much cheaper laptop does the same job.
Common questions
When did RTX Spark laptops come out? They went on sale worldwide on October 16, 2026, after pre-orders opened on October 7, according to NVIDIA and Microsoft.
How much does an RTX Spark laptop cost? NVIDIA says prices start at $2,599. Configurations with more memory cost more, and prices vary by region.
Is RTX Spark an Arm chip? Yes. Its CPU uses Arm cores, so Windows runs x86 apps through Microsoft's Prism emulator, while native Arm64 apps run directly.
Can RTX Spark run Fortnite and other online games? NVIDIA lists Fortnite, Counter-Strike 2 and Marvel Rivals among launch games, but some anti-cheat systems are not supported yet, and EA Javelin support is due in early 2027.
How do I check which GPU my current PC has? Open What GPU Do I Have? in your browser, or on Windows open Task Manager's Performance tab.