RTX Spark Linux bug caps GPU at 35% speed
Original: Nvidia RTX Spark encrypted firmware slows inference speed to 35% on Linux
Why This Matters
AI developers targeting Linux-first tools lose 65% GPU performance without a fix.
Nvidia's RTX Spark laptops running native Linux lock the GPU at ~780 MHz vs. 2.2 GHz on Windows, dropping inference from 65 to 14 tokens/sec on a 120B model due to encrypted firmware.
MindStudio's hands-on test of the MSI Prestige N16 Flip AI Plus — an Nvidia RTX Spark laptop — confirms Linux dual-boot works, but with a punishing performance catch. When plugged into wall power on Linux, the GPU clock speed collapses to roughly 780 MHz. On Windows, the same chip boosts past 2.2 GHz. The culprit: a Windows-only component called Nvidia Platform Controllers and Framework that manages CPU-GPU power delivery. Its firmware power table is encrypted, and only the Windows driver can decrypt it. Linux gets none of that.
The real-world impact is stark. Running GPT-OSS-120B, Windows hit 65.4 tokens/sec, WSL matched it at 65.1, and even Nvidia's own DGX Spark desktop managed 61. Native Linux on the same laptop: 13.9 tokens/sec — barely 21% of Windows throughput.
Memory allocation adds another wrinkle. The laptop reserves 96GB of unified memory for the GPU by default, and only Windows exposes the toggle to change that split. Without it, Linux sees only a fraction of the available RAM.
WSL is currently the practical workaround. It runs on the Windows driver stack, retains full GPU boost, and still supports Linux-first AI tools like vLLM. A permanent fix requires Nvidia to port the platform controller's power management logic into its Linux driver — no timeline announced.