RTX Spark Live Demos: See NVIDIA's Laptop Super-Chip In Action

NVIDIA made a big splash at Computex when it announced RTX Spark, an Arm-based AI super-chip designed in collaboration with MediaTek and Microsoft to power next-generation Windows systems. Or to put it in NVIDIA's own words, a chip designed to "reinvent Windows PCs for the age of personal AI." It's a big ambition for sure, and we had a chance to see RTX Spark up close and in action through a series of live demos performed on several Microsoft Surface Laptop Ultra systems equipped with the chip. 

Formerly codenamed N1 and N1X, RTX Spark combines NVIDIA's Blackwell-based GPU architecture with a 20-core Grace CPU, an NVLink C2C (chip-to-chip) interconnect, and up to 128GB of high-speed unified memory. Through key partnerships, NVIDIA is hoping to redefine the boundaries of mobile computing for the agentic AI era.

Don't mistake that to mean it's designed to only power glorified ChatGPT PCs. The first demo we saw highlighted the chip's gaming capabilities running Alan Wake 2 with full path tracing. Crucially, the demo also provided a glimpse of NVIDIA's upcoming DLSS 4.5 Ray Reconstruction update that is on track to arrive in August. It's impressive to see RTX Spark already running DLSS 4.5 Ray Reconstruction at this early stage, though the other major takeaway from that part of the demo is that NVIDIA's super-chip is going to support all of the latest RTX technologies right out of the gate.

The second demo on a separate RTX Spark-powered Surface Laptop Ultra gave us a look at the system handling an intricately detailed Unreal Engine city project loading entirely into the active memory pool. In a normal laptop environment, game developers are usually restricted by the amount of VRAM available to them, which sometimes means having to build environments in smaller, unlit blocks to prevent system crashes and hiccups.

For this demo, however, the system loaded 80GB of data—the entire Unreal Engine city—into unified memory in real time, a feat that is not feasible on traditional consumer mobile GPUs. RTX Spark enables developers to work inside a fully it, high-fidelity world without the typical constraints that would make this challenging.

The final demo shifted the focus towards on-device generative AI. It leveraged the super-chip's large pool of unified memory to run a heavy 35-billion-parameter Qwen 3.635B model that was running locally on 60-70GB of memory. RTX Spark is able to offload some of the repetitive development Q&A tasks that can bog projects down.

It will take some time to see what kind of impact RTX Spark can and will have in the PC space. What we saw is certainly impressive. Between the big pool of unified memory, next-gen DLSS 4.5 capabilities, and robust architecture partnerships, NVIDIA's ambitious goal looks like an attainable one.