Nvidia took the stage at Computex 2026 for its highly awaited entry into the consumer chip market – meet the RTX Spark, a chipset that merges a powerful and efficient ARM CPU with an RTX GPU.
The RTX Spark superchip merges an Nvidia Grace CPU with up to 20 cores with an Nvidia Blackwell GPU with 6,144 CUDA cores, and up to 128GB of fast, LPDDR5X unified memory.
The processor is based on Nvidia’s GB10 chip inside the DGX Spark platform, a class of enterprise-aimed mini-PCs running a custom Ubuntu Linux. The GPU is basically an RTX 5070 (non-Ti), having the same amount of CUDA cores. The chip uses TSMC’s 3nm node.
While the RTX Spark is undoubtedly ideally suited for content creation and gaming computers, Nvidia CEO Jensen Huang said the focus is on running AI agents locally on your computer. AI agents are the next step from LLMs (large language models), and can perform complex generation on demand. They require both powerful processing and quite a bit of memory, which is where the 128GB of unified memory comes in.
The RTX Spark is aimed at slim and powerful laptops, which are coming in the fall. Nividia has partnered with Acer, Asus, Dell, Gigabyte, HP, Lenovo, MSI, and Microsoft, which will introduce its most powerful laptop yet, the Surface Laptop Ultra, with the SoC.
There’s no word yet on product pricing, but it will be expensive. For context, Nvidia’s DGX Spark costs between $3,500 and $4,700, and can be configured with up to 128GB of RAM. Add a premium laptop with an OLED display, and you’ve got an eye-watering bill.
