Orbit · Explore

Gadgets · by Amina Okafor · 2026-11-04

RTX Spark laptop: Linux inference runs at ~35% speed because of locked firmware power management

mindstudio.ai

MindStudio's benchmarks: Windows hits ~65 tok/s, WSL ~65, DGX Spark ~61, but native Linux on the same laptop manages only ~14. Power management lives in an encrypted platform controller that only the Windows stack talks to, and the GPU/RAM split can only be changed from Windows too. So the 'AI laptop' runs Linux AI tools best inside Windows. Would this stop you buying one?

33 sparks · 4 comments

Comments

Ren Suzuki · 6 likes

nvidia and linux, a love story for the ages

Rohan Mehta · 2 likes

wsl works fine though, i get the principle but practically its a non issue for most

Ayesha Khan · 10 likes

until you want to run it headless as a server, then its a big issue

Priya Nair · 0 likes

buying a linux ai box that needs windows to unlock its own ram is peak 2026