Nvidia RTX Spark encrypted firmware slows inference speed to 35% on Linux

Native Linux on Nvidia RTX Spark laptops is up to five times slower when plugged in, limiting AI development.

Nvidia's RTX Spark laptops, designed for AI development with unified memory, face a significant performance bottleneck when running native Linux distributions while plugged into a power source. Benchmarks show GPU inference speeds dropping by as much as 75% on Linux compared to Windows or WSL (Windows Subsystem for Linux). This slowdown occurs because a proprietary Nvidia firmware component, responsible for managing power delivery between the CPU and GPU, uses encrypted power tables that only the Windows driver can decrypt. Consequently, the GPU clock speed on Linux is artificially capped at around 780 MHz when plugged in, compared to over 2 GHz on Windows, regardless of thermal conditions. While unplugged, Linux performance approaches Windows levels. Developers seeking full GPU acceleration on these machines currently find WSL to be the most viable option, as it leverages the Windows driver stack and avoids the Linux firmware's limitations.

The core issue lies in Nvidia's proprietary power management firmware, which uses encrypted tables inaccessible to Linux drivers. This prevents the RTX Spark GPU from reaching its full potential clock speeds when the laptop is connected to wall power, a critical scenario for sustained AI workloads. Developers are thus forced to choose between the convenience of Linux-first tools and native performance, or the higher speeds offered by Windows or WSL, which effectively run on top of the Windows driver stack.

This situation highlights a growing divide in AI development environments, where hardware vendors' proprietary software stacks can create performance disparities across operating systems. A true fix necessitates Nvidia porting its power management logic to its Linux driver, effectively bringing the DGX Spark's capabilities to a portable form factor. Until then, WSL will remain the pragmatic, albeit indirect, solution for developers who require both Linux tools and unfettered GPU performance on these high-end laptops.