Imagine buying a high-end AI powerhouse like the RTX Spark, only to find that switching to Linux—the gold standard for AI development—effectively cuts your hardware's brain in half. That is exactly the nightmare currently unfolding for developers and enthusiasts.
The 65% Performance Hit
Recent benchmarks reveal a staggering discrepancy in LLM inference speeds. While Windows and WSL (Windows Subsystem for Linux) can push GPT-OSS-120B to roughly 65 tokens per second, native Linux installs plummet to a meager 14 tok/s. That is a performance drop of roughly 65%, leaving users with only 35% of the speed they paid for.
The Culprit: Encrypted Firmware
The bottleneck isn't the driver, but a hidden power management controller. On Windows, the GPU clocks soar past 2 GHz, but on Linux, encrypted firmware locks the clocks at a dismal 780 MHz the moment the laptop is plugged in. Because this firmware is proprietary and encrypted, the Linux community is currently locked out of the logic required to unlock the full potential of the Tensor cores.
A Waiting Game
Currently, there is no official workaround. Until Nvidia releases the necessary logic for Linux drivers or the community finds a way to bypass the encrypted power limits, the RTX Spark remains a "Windows-only" beast for AI work. For a company that powers the AI revolution, this lack of Linux parity is a bitter pill for the dev community to swallow.
Sources
Media
