Nvidia’s RTX Spark laptops are built for a specific strength: running demanding AI workloads locally. PCWorld tested several models, including Microsoft’s Surface Laptop Ultra, HP’s OmniBook Ultra and a $7,000 ASUS ProArt P16 with 128GB of unified memory. The tests covered everyday tasks, games and local AI models, including OpenAI’s 120-billion-parameter gpt-oss.
The platform combines a 20-core Arm CPU and a Blackwell integrated GPU with a shared pool of memory that can reach 128GB. Unlike a conventional gaming laptop, which has a smaller amount of memory dedicated to its graphics card, RTX Spark lets the CPU and GPU access the same pool. On the ProArt, roughly 96GB could be assigned to the GPU, making it possible to run models that exceed the memory available to many conventional systems.
In one test, the ProArt processed 472,084 tokens with the gpt-oss model in 33 minutes and 13 seconds. A desktop equipped with a GeForce RTX 5090 took 52 minutes and 51 seconds on the same workload. PCWorld noted that the laptop’s GPU is not generally faster: the result reflected the unusually large amount of GPU-accessible memory required for that particular task. When a workload fits within the RTX 5090’s 32GB of graphics memory, the desktop can make better use of its faster GPU.
That makes RTX Spark most relevant to developers and researchers who run large models on their own machines, or who want to avoid paying for cloud inference and keep data local. For smaller local models or ordinary use of online AI services, the extra capacity may not be necessary. The review cautioned that buyers considering a high-priced configuration should have workloads that can take advantage of its memory.
Gaming is possible, too. PCWorld’s tests found the integrated Blackwell graphics roughly comparable to a GeForce RTX 5070, with Nvidia features such as DLSS upscaling and frame generation available in supported games. However, conventional GeForce laptops delivered better gaming performance for less money, and the additional unified memory did not improve game results. The review therefore treats gaming as a useful extra rather than a reason to buy an RTX Spark machine.
For routine use, these laptops behave much like other Windows 11 PCs. The Surface Laptop Ultra delivered strong multicore performance, though its single-core results were less impressive. Its battery lasted about 10 hours in PCWorld’s office-work testing. Heavy AI work is more demanding: during one ProArt run, the fans became loud and the laptop used 46 percent of its battery in 33 minutes.
The Arm processor architecture is another consideration. Microsoft’s Prism translation layer supports a wide range of traditional x86 Windows software, and native Arm support is expanding, but compatibility is not universal. During testing, PCWorld encountered software issues, including the need to use a beta version of Adobe Premiere because the released version did not properly support the platform’s GPU. Buyers who rely on specialized programs, drivers or plug-ins should check compatibility before purchasing.
PCWorld’s assessment is that RTX Spark is aimed at a relatively small group: people with substantial local AI workloads, including some creative professionals whose essential software works on the platform. Gamers and buyers seeking a premium everyday laptop can find more suitable options for less. RTX Spark’s appeal lies in its ability to bring large amounts of shared memory and serious local AI computing to a portable PC, not in being the best-value choice for everyone.
Sources: PCWorld.