By Hamza Faisal

Nvidia's RTX Spark laptops were built to put a full AI workstation in a bag. On Linux, they are running at a fraction of the speed buyers paid for — and the reason, according to an independent test, appears to be locked inside the firmware itself.

MindStudio, an AI research group, put an MSI Prestige N16 Flip AI Plus through a dual-boot experiment and published the results on November 1. The laptop carries Nvidia's GB10 chip and 128GB of unified memory, the DGX Spark desktop's silicon. Running the open-weight GPT-OSS-120B model, it managed 65.4 tokens per second on Windows — and just 13.9 on native Linux while plugged in. Under Windows Subsystem for Linux, the same model hit 65.1 tokens per second, which rules out the hardware as the problem.

The numbers: same laptop, same model, very different speeds

The gap showed up in every test. With vLLM serving eight simultaneous clients, the laptop reached 127 tokens per second under WSL and only 39.5 on native Linux. The GPU clock told the same story: roughly 2.2 GHz on Windows, but pinned near 780 MHz on Linux while plugged in. Oddly, Linux ran faster on battery than plugged in. Nvidia's DGX Spark desktop, which ships with Nvidia's own Linux build, ran the same model at 61 tokens per second without any of this friction. The lock sits on the laptop side of the GB10 line, not the chip itself.

Why Linux can't fix it — the power table is encrypted

MindStudio's explanation is blunt: the firmware's power table is encrypted, and only Nvidia's Windows driver can decrypt it. The component in question is Nvidia Platform Controllers and Framework, which manages power delivery between the CPU and GPU. On Linux there is no equivalent mechanism, so the chip never gets permission to run at full speed.

The group tested that theory with a control experiment. Disabling Platform Controllers and Framework in Windows dropped throughput from about 83 tokens per second to 31 and reproduced the exact 786 MHz clock seen on Linux; re-enabling it restored full speed instantly. MindStudio reports that no software-level fix on Linux worked, and the findings reached the Hacker News front page on November 4, collecting 555 points and 253 comments. Nvidia has not responded publicly, and its open-gpu-kernel-modules tracker still carries an open GB10 issue filed on September 13.

What this means for buyers

RTX Spark laptops are sold as portable AI machines, and the pitch only works if the silicon runs at its rated speed. The affected line is not a single model: besides MSI's Prestige, RTX Spark systems are shipping from Asus under the ProArt name and from Dell under XPS. Anyone planning to run local models on native Linux is, for now, buying a machine that performs like a much cheaper one the moment it is plugged in.

Two caveats are worth stating plainly. This is one machine and one model, tested by one group, and the figures are MindStudio's measurements rather than independently verified benchmarks. The often-quoted 35 percent figure matches no calculation in the data, which works out to roughly 21 percent on the main test and 31 percent on the vLLM test. But the direction is consistent across every measurement, and the Windows control experiment is hard to argue with. Until Nvidia ships a Linux driver that can read its own power table — or unlocks the firmware — RTX Spark owners who live on Linux have three options: run Windows, run WSL, or accept a laptop that runs at a third of its speed.