Nvidia's RTX Spark runs inference at 35% speed on Linux, a slowdown traced to encrypted firmware on the device, according to a discussion that gained traction on Hacker News on November 4, 2026. The thread treats the number as a baseline complaint rather than a one-off bug report, with developers comparing notes on a machine marketed in part for local model work outside Windows.
RTX Spark belongs to a category of compact Nvidia hardware built to run models on a desk instead of in a data center. Builders chose this class of device specifically to avoid cloud inference costs and latency, and Linux is the default environment for much of that workflow, from containerized serving stacks to custom CUDA builds. A firmware layer that behaves differently depending on operating system cuts against that expectation, since the hardware's appeal rests on predictable, portable performance.
The specific mechanism behind the slowdown, whether it stems from thermal management, driver scheduling, or something else encoded in the encrypted firmware itself, was not disclosed in the discussion. What is documented is the outcome: a 35% speed figure on Linux, reported by the community rather than by Nvidia. That gap between vendor silence and user-measured performance is itself the story, since encrypted firmware means outside developers cannot inspect what is throttling the chip or confirm whether the behavior is intentional.
For anyone evaluating RTX Spark for a Linux-based inference pipeline, the immediate step is benchmarking against the specific model and batch size planned for production, not trusting a spec sheet number. Teams that budgeted for near-parity with Windows performance should revisit those assumptions before committing rack space or procurement dollars. Encrypted firmware also means there is no clear path to a community patch or workaround, the kind that has historically rescued other hardware from vendor neglect, so the unresolved question is whether Nvidia addresses this through an update or leaves Linux as a second-tier target indefinitely.
The pattern echoes earlier friction between Nvidia and the Linux community, where driver quality and feature parity have lagged Windows releases for years even as Linux became the default OS for serious machine learning infrastructure. What is different here is the opacity: previous complaints centered on open drivers that could at least be examined or patched, while encrypted firmware forecloses that option entirely.
Nvidia has not issued a public response addressed in the discussion, leaving builders to treat the 35% figure as the working number until the company says otherwise.