Last week, I spent two days exploring how Microsoft and NVIDIA are redefining personal computing with RTX Spark, a new system-on-chip that powers the Surface Laptop Ultra and several other Windows PCs. This is the first time a powerful Arm CPU has been integrated with NVIDIA's Blackwell GPU technology and up to 128GB of ultra-fast unified memory on a single chip. In theory, RTX Spark should enable thin yet powerful laptops (and several compact desktops) capable of handling AI tasks locally, as well as content creation and gaming.
Therefore, I think NVIDIA and Microsoft have reason to feel a bit proud. However, as was evident when the Snapdragon-powered Copilot+ system was released two years ago, it cannot be ignored that RTX Spark is clearly a response to Apple Silicon, which brought transformative changes to the Mac six years ago. Apple's chips also consist of powerful Arm CPUs, capable GPUs, and fast unified memory, covering a full range of products from the compact Mac mini to the powerful Mac Studio and 16-inch MacBook Pro.
With the launch of RTX Spark, Windows users finally have the opportunity to experience the advantages of efficient Arm-based system-on-chips in high-end personal computers, while the latest batch of Qualcomm chips covers the low-end and mid-range markets. There were not many surprises at Microsoft's AI launch event (which showcased the Surface Laptop Ultra) and NVIDIA's RTX Spark editorial day, which I attended along with other media and analysts. Both companies aimed to show us a future trend where AI models and agents can run directly on RTX Spark-powered Windows laptops, a point that has been clear since the chip was announced at Computex.
Even if such a future may not materialize, RTX Spark could still bring some noteworthy changes to the personal computer landscape.
Performance and Efficiency of the RTX Spark System Will Change the Personal Computer Market
The RTX Spark SoC features up to 20 NVIDIA Grace Arm CPU cores (10 performance cores and 10 efficiency cores), a Blackwell GPU with up to 6,144 CUDA cores, and up to 128GB of unified LPDDR5x memory running at 9,400 Mbps. There is also a slightly weaker configuration with 18 CPU cores and 5,120 GPU cores.
RTX Spark also includes many features from NVIDIA's latest GPUs, including fifth-generation tensor cores, fourth-generation ray tracing cores, and the Blackwell video engine (supporting AV1 UHQ and 4:2:2 encoding and decoding). NVIDIA claims that by integrating all this hardware onto a single chip, RTX Spark's efficiency is twice that of a laptop equipped with an Intel Core Ultra 9 285H and an RTX 5070 discrete GPU, consuming only 60 watts during gaming.
NVIDIA previously stated that the performance of the RTX Spark system should be comparable to that of the RTX 5070 laptop, although the following chart shows that those discrete GPU systems also consume significantly more power for better performance.
If you plan to stay plugged in most of the time and do not need to carry your laptop around, choosing an RTX Spark system may not offer many advantages. However, if you need to do serious creative work on the go, according to NVIDIA, the improved efficiency of RTX Spark allows PCs to perform at speeds comparable to when plugged in while on battery power. Based on my brief experience with the Surface Laptop Ultra, I did not notice any performance drop while running Gears of War: E-Day on battery power.
The game remained smooth at 1440p high-quality settings. With support for a large amount of unified memory, the RTX Spark system should be attractive to AI developers and researchers who require substantial memory to run large AI models locally, such as Qwen3.8-Flash-Next with 125B parameters. The RTX Spark system could significantly reduce spending on cloud AI services or completely replace that need. I spoke with a researcher who envied colleagues' ability to easily run AI models on modern Mac laptops and was excited about achieving the same functionality on a Windows laptop.
The launch of RTX Spark coincides with Microsoft's addition of support for AI tools (like Perplexity) on devices in Windows 11, as well as simplifying the setup of AI agents (like OpenClaw and GitHub Copilot). Undoubtedly, NVIDIA is also promoting the gaming capabilities of RTX Spark. With its Blackwell GPU, it will provide all the features you expect from RTX 50 series graphics cards, including support for DLSS 5.
As I mentioned earlier, you can expect performance similar to that of the RTX 5070, and NVIDIA specifically claims that 1440p is the expected resolution for the RTX Spark system, although I am also curious to see how they perform at 4K resolution with DLSS. The new performance mode for DLSS frame generation can allow Cyberpunk 2077 to reach frame rates of up to 280fps using 6x frame generation, while the non-performance mode yields 213fps. (However, it remains unclear what sacrifices are made to achieve this frame rate boost.)
Since RTX Spark is based on Arm architecture, it will not natively run all games on Windows, but NVIDIA claims that recent games like Control Resonant and the aforementioned Gears of War: E-Day will be supported at launch, thanks to Microsoft's Prism emulator for x86 software. Major DLSS 5 launch titles like NBA 2K27 will also be supported, along with World of Warcraft: Forever.
NVIDIA stated they are also working with major anti-cheat vendors to support their products, which has previously hindered multiplayer gaming on Arm Windows systems.
I need to spend more time with the RTX Spark system to understand its actual gaming performance beyond my few minutes with Gears of War: E-Day. However, if gaming is your primary consideration, you would still be better off choosing an Intel or AMD system with a discrete GPU, so you don't have to worry about potential emulation issues. The RTX Spark system is also useful for content creation—something you would expect from a PC equipped with an RTX Blackwell GPU.
You can rely on hardware acceleration in tools like Blender, and it is worth noting that Adobe is specifically restructuring Photoshop and Premiere for NVIDIA's platform. This is a significant show of confidence from the major software company, although I think Adobe is also looking ahead, recognizing the bright future of Arm-based Windows systems.
NVIDIA does not intend to target the RTX Spark at ordinary consumers, as the starting price for the 18-core system is US$2,600 (approximately HK$20,280), while the full 20-core model goes up to US$3,700 (approximately HK$28,860). NVIDIA and Microsoft's target audience is AI developers and researchers who can afford to spend significantly on large amounts of RAM. They should be prepared to open their wallets—the Surface Laptop Ultra with 128GB RAM will be sold for US$5,900 (approximately HK$46,020).
However, this is still much cheaper than the fully configured 16-inch MacBook Pro with M5 Max (128GB RAM priced at US$7,000 (approximately HK$54,600). The mere existence of the RTX Spark platform may push Intel and AMD to innovate in new directions. AMD is attempting to achieve similar goals with its Ryzen AI Max 400 chips, which can also accommodate large amounts of unified memory but can run existing x86 software without the need for emulation.
Meanwhile, Intel has been eager to prove its competitiveness with AMD in recent years, leading to the creation of its powerful Core Ultra series 3 chips. Competition is always a good thing for consumers.
Item Specifications Processor Up to 20 NVIDIA Grace Arm CPU cores GPU Blackwell GPU, up to 6,144 CUDA cores Memory Up to 128GB LPDDR5x, speed 9,400 Mbps
Disclosure: This article contains affiliate product links, and if you purchase through these links, TechRitual may earn a commission, but this does not affect product reviews and recommendations. For more details, please refer to the privacy policy.