NVIDIA RTX Spark Specifications
GPU
Up to 6,144 CUDA
CPU
Up to 20-Core Grace
Memory
Up to 128GB Unified
Standard SKU: 18-core CPU + 5,120-core GPU (up to 64GB)
Specifications compiled from NVIDIA official documentation. Process node and clocks are based on published architecture whitepapers.
The NVIDIA RTX Spark is not just a new laptop chip; it is a "superchip" designed to bring enterprise-level AI to a portable Windows machine. By combining the 20-core Grace CPU and the Blackwell RTX GPU into a single package, NVIDIA claims it can run 120-billion-parameter AI models locally.
But independent benchmarks tell a more nuanced story. The AI performance is genuinely class-leading. The gaming performance is more complicated. Here is the fact-checked breakdown of its real-world performance, AI capacity, and exactly where you can buy it.
What Exactly is the RTX Spark? (CPU, GPU, or Both?)
To answer the most common question first: The RTX Spark is both a CPU and a GPU. It is a System-on-a-Chip (SoC).
Traditionally, a laptop has a CPU for general tasks, a separate GPU for graphics and AI, and a motherboard connecting them to the RAM. The RTX Spark changes this by putting the CPU, the GPU, and the memory onto a single piece of silicon, connected by an NVLink-C2C interconnect running at 600 GB/s.
If you are familiar with Apple's M-series chips or Qualcomm's Snapdragon X Elite, the RTX Spark uses the same architectural philosophy—but with an NVIDIA Blackwell GPU attached to it.
Gaming Performance: What FPS Can It Actually Play?
This is where the marketing claims and independent benchmarks diverge. NVIDIA's official press release claims "AAA games at 1440p at over 100 frames per second" — but that figure depends entirely on DLSS 5 and frame generation being enabled.
Independent testers have now published real numbers. Here is what they measured, not what was promised:
- Cyberpunk 2077 (1440p, Ray Traced Ultra, Frame Generation ON): ~60 FPS on the Surface Laptop Ultra.
- Cyberpunk 2077 (1080p, Extreme preset, RT ON, No DLSS): 39 FPS. With DLSS enabled: 80 FPS.
- Gears of War: E-Day (2180×1453, Ultra, DLSS Dynamic, Mega Geometry): 123 FPS average — but the 1% low was just 28 FPS, with visible microstuttering.
- Black Myth: Wukong (1080p, Cinematic, No DLSS): 36 FPS.
- Forza Horizon 6 (1440p, Extreme, RT ON, No DLSS): 29 FPS average, 15 FPS 1% low. DLSS is mandatory for playable framerates.
Compatibility is also a concern. Red Dead Redemption 2 crashed when launched with Vulkan, only working after switching to DirectX 12. Games with strict kernel-level anti-cheat — like Valorant — may not run at all on Arm architecture.
The honest summary: the RTX Spark delivers RTX 5070-class graphics performance, but with higher latency and more compatibility caveats than a traditional x86 gaming laptop.
Local AI Capacity: What Can It Run Simultaneously?
This is the RTX Spark's primary selling point, and the benchmarks here are genuinely impressive. The 128GB unified memory pool means you can load massive models entirely on-device, avoiding cloud API costs and data sovereignty concerns.
- Maximum Model Size: Up to 120 Billion parameters with 1M token context.
- AgentPerf Test (Qwen 3.8 27B): The Surface Laptop Ultra completed the task in 18.9 minutes. A 14-inch MacBook Pro with M5 Max took 24.74 minutes.
- Less Demanding Test (Qwen 3.6 35B A3B): 8 minutes on RTX Spark vs. 10 minutes on M5 Max and 13 minutes on AMD Ryzen AI Max+ 395.
- Simultaneous Workloads: NVIDIA claims the shared memory allows 90GB+ 3D scene rendering, local AI agent execution, and Adobe Premiere video editing to run concurrently.
Creator & Developer Features
Beyond AI and gaming, the RTX Spark is explicitly designed for professional creators and software developers. NVIDIA has engineered the chip to handle workflows that traditionally require dedicated workstation hardware.
- Full 4:2:2 Video Support: Video editors can work with 4:2:2 color subsampling in full quality without transcoding, which is a massive time-saver for professional workflows.
- Hardware Codecs: Dedicated hardware encode and decode for HEVC, H.264, and AV1.
- NVIDIA Broadcast & RTX Video: Real-time AI enhancements for streamers and content creators, including background blur, noise removal, and video upscaling.
- Developer Stack: Native support for the full CUDA stack, TensorRT, and NVFP4.
Test it yourself below — will your model fit in the RTX Spark's memory?
Interactive: Will it run your model?
Select a model size to see if it fits in the RTX Spark's memory (including OS overhead):
Every Laptop with the NVIDIA RTX Spark (2026 Lineup)
Because NVIDIA supplies the superchip to third-party manufacturers, there is no single "NVIDIA Laptop." Instead, six major brands are implementing the RTX Spark across eight distinct models. Here is the complete list, based on current retail listings.
| Laptop Model | Starting Price | Max Memory |
|---|---|---|
| Microsoft Surface Laptop Ultra | $2,599 | 128GB |
| ASUS ProArt P16 | $4,399 | 128GB |
| ASUS ProArt P14 | $3,699 | 64GB |
| Dell XPS 16 Creator Edition | $3,799 | 64GB |
| HP OmniBook X 14 | $2,999 | 64GB |
| HP OmniBook Ultra 16 | $3,199 | 128GB |
| Lenovo Yoga Pro 9n | TBA | 128GB |
| MSI Prestige N16 Flip AI+ | TBA | 64GB |
Where to Buy (Regional Availability)
Because it uses an Arm-based architecture, you should verify that your specific software runs natively or via Microsoft's Prism emulator before buying.
North America
- All eight models are available for pre-order or direct purchase through Microsoft, ASUS, Dell, HP, Lenovo, and MSI online stores.
India
- Microsoft Surface Laptop Ultra: Listed at ₹2,49,999 for the 20-core CPU model. Availability is marked as "Upcoming".
- ASUS ProArt P16: Listed at ₹1,99,999 for the 24GB / 512GB model. Higher-memory configurations will cost more.
- Note: These are early listings and may shift before official retail availability. Import duties and regional taxes can add 15-30% to US pricing.
Europe / UK
- Pre-orders opened on October 7, 2026. The Surface Laptop Ultra starts at approximately €2,900 for the 24GB model. Dell, HP, and Lenovo are selling through business channels.
The Bottom Line
The RTX Spark is the most ambitious Windows chip released this year. NVIDIA claims it is the "most power-efficient RTX chip ever made," designed to deliver all-day battery life without sacrificing compute power. If you are a developer paying for OpenAI API tokens, or a creator working with massive 3D scenes, the unified memory pays for itself quickly. The AI benchmarks are not marketing — they are real, and they beat the M5 Max in agentic tasks.
But if you just want to play games, the picture is murkier. The RTX Spark delivers RTX 5070-class performance, but with higher latency and more compatibility caveats than a traditional x86 laptop. A conventional RTX 5070 laptop will offer better raw gaming performance for the money.
For a machine that does both AI and gaming on the go, the RTX Spark is currently in a class of its own. For everything else, the competition is closer than the spec sheet suggests.
Frequently Asked Questions
Is RTX Spark better than an RTX 5070 laptop?
In gaming, no. Independent benchmarks show it performs at roughly RTX 5070 laptop levels, but with higher latency when frame generation is enabled. In AI, yes — it offers up to 128GB of unified memory, which no dedicated RTX 5070 laptop has.
Can I upgrade the RAM on an RTX Spark laptop?
No. The LPDDR5X memory is soldered onto the superchip package. You must choose your memory configuration at the time of purchase.
Does RTX Spark run Windows?
Yes, it runs Windows 11 on Arm. NVIDIA has worked with Microsoft to ensure day-one support for local AI agents and major creative applications, including Adobe Photoshop and Premiere.
What is the battery life like?
NVIDIA claims "all-day battery life" due to the 45W-80W TDP and the efficiency of the Grace CPU. Independent battery testing has not yet been published. Heavy gaming or AI workloads will drain the battery faster than light productivity tasks.
Can I run games with anti-cheat on RTX Spark?
Some games, yes. Others, no. Fortnite and Marvel Rivals are reported to work. Games with strict kernel-level anti-cheat, like Valorant, may not run on Arm architecture at all. Check compatibility before buying if competitive gaming is your priority.
Sources & Fact-Check
This article was drafted with AI assistance and reviewed by a human editor. All specifications, pricing, and performance figures were cross-referenced against the following primary sources:
- NVIDIA Newsroom — RTX Spark PCs Available Today (October 16, 2026)
- NVIDIA — RTX Spark Product Page
- PCMag — RTX Spark Laptop Preorders Begin (October 7, 2026)
- Mezha — NVIDIA RTX Spark and DGX Station Pricing
- Mobidevices — RTX Spark Laptops Tested in Games and AI Workloads (October 16, 2026)
- VGTimes — RTX Spark Excels at Local AI but Struggles in Gaming Tests (October 15, 2026)
- Smartprix — NVIDIA CPU Laptops Price List (India)