Eigen RadarAI
Analysis

RTX Spark’s local AI tasks consume battery and need corrections

A week of hands-on testing found that running AI locally on RTX Spark laptops could take substantial battery power and repeated prompts. A news briefing and a five-image editing job exposed different limits in the tested workflows. Windows offered simple installation options, but choosing a suitable model and handling compatibility still required work.

Artificial Intelligence··Night
Rear view of an open laptop connected to its power adapter.

News gathering drains the Asus battery

A daily news briefing became a ten-minute local task in Sean Hollister’s RTX Spark laptop trial. Running Qwen3.8-27B through Hermes, software for carrying out AI tasks, used 10 percent of the Asus battery. In his comparison, cloud-based Claude Opus 5.5 finished in about half the time, with detail and sourcing he preferred. Those results describe his particular workflow.[1]

After reinstalling Hermes, Hollister spent longer on another attempt, using 60 percent of the battery. The test was part of a week spent with NVIDIA’s Arm-based N1X laptops by three Verge reviewers.[1]

Five images need another round of editing

Antonio G. Di Benedetto asked the model to resize five product images and give them white backgrounds on a Lenovo laptop. After 36 minutes and 30 percent battery use, the products were incorrectly cropped. Di Benedetto corrected the result through additional instructions while the laptop drew mains power.[1]

Easy installation leaves model selection to users

Windows 11 lets users install tools including Perplexity, Hermes and OpenClaw from an introductory screen. With less memory available, Hermes suggested cloud options; using a smaller local model instead required users to select one themselves. Tom Warren described an experience still resembling early developer tools. Ordinary applications generally worked for Hollister through Windows’ Prism compatibility layer, although gaming limitations remained.[1]

References

  1. News sourceThe VergeRTX Spark tests find local AI brings battery and correction costs↩1↩2↩3↩4