Hardware benchmarks published by MindStudio evaluating MSI Prestige N16 Flip AI Plus laptops powered by the Nvidia RTX Spark mobile architecture revealed severe performance degradation when running native Linux distributions.
While Linux installs successfully and leverages Nvidia's open-source kernel modules for desktop acceleration, plugging the workstation into AC wall power locks the GPU core frequency to approximately 780 MHz. In benchmark tests evaluating inference speed on GPT-OSS-120B, native Linux achieved only 13.9 tokens per second, compared to 65.4 tokens per second on Windows 11, 65.1 tokens per second on Windows Subsystem for Linux (WSL), and 61 tokens per second on the dedicated desktop DGX Spark.
The investigation traced the bottleneck to the Nvidia Platform Controllers and Framework (NPCF), a Windows-exclusive subsystem responsible for negotiating dynamic thermal design power (TDP) sharing between CPU and GPU. Under Windows, NPCF dynamically boosts the GPU to 2.2 GHz, but its internal power curve tables are encrypted and lack a functional Linux implementation.
Memory allocation presents an additional structural challenge for Linux users: the hardware automatically partitions 96 gigabytes of unified memory to the GPU by default, leaving insufficient RAM for system processes unless adjusted using a proprietary Windows control utility before booting Linux.
Consequently, researchers and developers seeking to run heavy local models on RTX Spark laptops are currently forced to utilize WSL2 inside Windows or accept dramatic compute penalties until Nvidia provides an open Linux power management implementation.