What was announced
Nvidia announced RTX Spark on May 31, describing a superchip for Windows laptops and compact desktops designed around local AI workloads, graphics and games. The company said systems would support up to 128GB of unified memory and identified PC partners including Microsoft Surface, Dell, HP and Lenovo, with initial products planned for autumn.
Nvidia also described work with Microsoft on Windows support for agents. Performance, battery-life and model-capacity statements in the release are manufacturer claims, not results from this publication’s testing.
What matters beyond the headline specification
A peak AI throughput figure does not tell a buyer how quickly a particular application will finish, how much memory remains available, or how a laptop behaves on battery power. Compare the complete configuration and the software you actually use.
For local agents, permissions and recovery matter alongside speed. Before allowing software to change files or act through other applications, establish approval steps, backups and a way to inspect its actions. Local execution is an architectural choice; it does not make every workflow reliable automatically.
Write a workload-based comparison
For an illustrative local-AI evaluation, give each candidate machine the same permitted model and input task. Record completion time, memory use, output quality, power mode and whether any step sends data to an online service. Note the operating-system and software versions so the result can be repeated.
A faster demonstration with a smaller model does not answer how the computer handles your larger workload. Likewise, a desktop plugged into mains power is a different comparison from a laptop running on battery. Define the conditions before comparing results, and include the cost of the configuration that actually passed the test.
