RTX Spark 体验：NVIDIA 终于将 RTX 游戏与本地 AI 带到 Windows on Arm

For the past five days, I’ve been using the ASUS ProArt P16 as my first proper experience with NVIDIA’s RTX Spark platform. It’s an unusual machine, which is part of what makes it interesting. If you don’t already know it yet, RTX Spark combines an Arm-based NVIDIA Grace CPU with a Blackwell GPU and unified memory, instead of pairing an x86 processor with a discrete graphics chip. ASUS offers the ProArt P16 with up to 128GB of unified memory and a 16-inch 4K OLED touchscreen, but the unit NVIDIA provided has 64GB of memory and a lower-resolution 3K OLED display. Those differences matter when assessing the laptop itself, but in this article I’m more interested in sharing what RTX Spark can do.

That combination is meant to handle creative work, local AI models and PC gaming in one relatively slim laptop. NVIDIA’s pitch includes running large AI models locally and playing demanding games with its RTX technologies, but I haven’t had enough time to test all of that properly, and NVIDIA’s review schedule has been tight.

Even in a short time, however, RTX Spark has made the ProArt P16 feel different from the Windows notebooks I’m used to reviewing. The processor architecture is Arm, but the graphics side is unmistakably NVIDIA: Blackwell, CUDA, ray tracing and DLSS 5 are all part of the package. This isn’t simply another thin Windows laptop with integrated graphics. It’s an attempt to bring Arm’s efficiency together with the graphics capabilities people expect from an RTX machine.

Windows on Arm is part of the experience

RTX Spark is built for Windows on Arm, and its arrival is significant for Qualcomm’s Snapdragon X Elite series, which helped bring Windows on Arm to a wider audience with laptops focused on everyday computing, battery life and on-device AI. RTX Spark brings NVIDIA’s graphics technology and local AI tools into the same Windows ecosystem, but with a different emphasis. That said, I don’t see RTX Spark making Snapdragon X Elite obsolete. The platforms target different priorities, but NVIDIA’s presence is more likely to encourage wider support for games and applications on Windows on Arm.

Beyond its CPU architecture, RTX Spark also joins a growing group of platforms built around unified memory. Apple has used a unified memory architecture in its Apple silicon Macs since the M1 arrived in 2020, and the latest MacBook Pro with M5 Max can be configured with up to 128GB. AMD has brought a similar approach to Windows PCs with its Ryzen AI Max+ series but unlike RTX Spark, Apple’s M-series and Qualcomm’s Snapdragon X Elite, the AMD’s platform is still based on x86.

The interesting part, then, is how RTX Spark combines shared memory with an Arm CPU, NVIDIA’s Blackwell graphics and RTX technologies, all running Windows. Apple has built its own hardware and software ecosystem around unified memory, while AMD offers an x86 Windows alternative with integrated Radeon graphics. RTX Spark brings a different mix of hardware and software to Windows laptops, and I think it’ll be interesting to see how future, lower-priced, RTX Spark generations of computers might make agentic AI coding (or AI coding in general) even more accessible to the mainstream audience.

Gaming, creating and learning with RTX Spark

Gaming gave me the quickest feel for RTX Spark, though I’ve also spent time editing in Adobe Premiere and trying a few simple AI inference tasks. I played Cyberpunk 2077 and Gears of War: E-Day at 1440p with mid-range settings. Cyberpunk 2077 held a fairly steady 75fps, and both games looked great on the ProArt P16’s 3K display. Do you really need a 4K screen for gaming on a laptop then? It’s your call, but I stand by my past statement that 4K gaming on a small laptop screen is a terrible waste of power.

What about thermals and power throttling? Well, compared with standard gaming laptops with discrete RTX 50 series GPUs that I’ve tested, the ProArt P16 also stayed more than comfortable to use during my extended play sessions too. Its keyboard and palm rest do become noticeably warm during longer sessions, but never to the point where it felt like your wrists were being cooked.

Then there’s content creation which I’m still getting used to editing in Premiere on this machine, so I need more time before I can say how it handles longer projects or video exports. But the AI side has been more immediately engaging for me because I’m new to large language models (LLMs) and AI coding, and trying a few simple inference tasks locally with tools such as Ollama and LM Studio has given me a hands-on way to learn how models work. It has also made me curious about how AI agents fit into this kind of workflow. Knowing that these particular experiments were running locally was reassuring (dependent on the model and features being used) and being able to explore these tools within the familiarity of Windows has been one of the more enjoyable parts of using RTX Spark.

An early impression

Taken together, gaming, Premiere and local AI have given me a clearer sense of what RTX Spark is trying to do, but I still have questions about how well its hardware and Windows software work together. Price adds another consideration. ASUS lists the ProArt P16 with RTX Spark from S$6,599 in Singapore, for a configuration with 48GB of memory and a 1TB SSD. The 64GB model costs S$7,499 (the version I tested), while the 128GB version is S$11,099. By comparison, Apple’s 16-inch MacBook Pro with M5 Max starts at S$5,449, with 48GB of memory and a 2TB SSD. At their starting prices, the ASUS costs more while offering half the storage – and we know how storage prices are nowadays. These aren’t matching configurations, of course, and the different chips and operating systems make this a price comparison rather than a performance one.

That premium may make more sense for a narrower group of buyers, however: developers who rely on CUDA and want to run AI models locally. RTX Spark combines NVIDIA’s software stack with configurations of up to 128GB of unified memory. Apple also offers up to 128GB on the M5 Max, so the difference lies in the tools, apps and workflows each platform supports. The extra cost is easier to justify if NVIDIA’s stack fits the work you need to do.

I’m still running tests and benchmarks, and I’ll cover performance in a separate feature before reviewing the ASUS ProArt P16 itself. But RTX Spark has also got me thinking about who might find a machine like this useful. Most laptops have limited graphics memory for local AI, and a laptop like the RTX Spark gives local models more room to run with its larger unified memory. For someone like me, who is new to LLMs and AI coding, tools such as the open-source Ollama make it easier to download a model and start experimenting without setting up everything from scratch. The ProArt P16 is expensive, for sure, so it won’t make local AI accessible to everyone. But if the same combination of memory and CUDA-powered software reaches more afforable computers in the future (and when the current RAMmageddon blows away), it could give more people a straightfoward way to try local models and AI agents.

RTX Sparks laptops are now available from NVIDIA partners, including ASUS, Microsoft, Lenovo, Dell, MSI and HP.