TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A MindStudio report found that native Linux on an MSI Prestige RTX Spark laptop delivered 13.9 tokens per second on GPT-OSS-120B, compared with 65.4 on Windows. The report links the gap to Linux’s inability to access a Windows-only power-management component and encrypted firmware tables; Nvidia has not provided a response in the supplied material.
A MindStudio report says native Linux ran GPT-OSS-120B at 13.9 tokens per second on an MSI Prestige RTX Spark laptop, compared with 65.4 on Windows, after the GPU stayed near 780 MHz while plugged into wall power. The report attributes the slowdown to a Windows-only Nvidia power-management component and encrypted firmware tables that Linux cannot currently read, a limitation that could matter to developers buying these laptops for local AI work.
In MindStudio’s reported benchmark, the same laptop reached 65.4 tokens per second on Windows and 65.1 in Windows Subsystem for Linux (WSL). Native Linux managed 13.9 tokens per second. The report also gives a comparison of 61 tokens per second for an Nvidia DGX Spark desktop, though the supplied material does not detail whether every test condition was identical across systems.
MindStudio says the laptop’s GPU ran at roughly 2.2 GHz under Windows but about 780 to 790 MHz under Linux when connected to power. In a longer run, Windows finished in about eight minutes, while Linux took 37 minutes, according to the report. It says temperatures remained between 40 and 47°C during the tests, pointing away from thermal throttling as the explanation. These are measurements from the report, not independently verified results in the supplied source material.
The report identifies Nvidia Platform Controllers and Framework as the missing link: a Windows-only component that manages power delivery between the CPU and GPU. It says Linux cannot access the relevant encrypted power table, leaving the GPU at a lower clock. A separate configuration issue also affects usable memory: the laptop reserves 96GB of unified memory for GPU use by default, and the report says changing that allocation requires Windows. After lowering the reservation to 4GB in Windows, MindStudio says Linux could see 120GB of the machine’s 128GB.
Linux Limits Local AI Throughput
The result matters because RTX Spark laptops are presented as machines for AI development, with unified memory configurations up to 128GB and Nvidia GPU acceleration. Many AI tools, including vLLM and SGLang, are built with Linux workflows in mind. A slowdown of the size reported could make native Linux a poor fit for sustained inference on this hardware, even when the operating system boots and the GPU is available.
That distinction is practical, not just a benchmark concern. Developers may need to choose between running Linux directly, using Windows with WSL, or waiting for driver support. MindStudio’s figures suggest WSL retained near-Windows performance in its tests, offering a way to use Linux-based tools while relying on the Windows driver stack. The report also cautions indirectly that booting successfully does not mean the system’s power controls and memory allocation work as expected under Linux.
The published results should not be treated as a universal measure for every RTX Spark laptop or Linux setup. The supplied report focuses on an MSI Prestige model and specific workloads. Still, it identifies a potential platform-level compatibility problem that could affect buyers who expect the same GPU performance across operating systems.
Popular right now
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Hazelnut Taupe
Dyson W1 PencilWash Wet and Dry Cleaner with Dok Cordless Wet Floor Cleaner
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Almond Cream
As an affiliate, we earn on qualifying purchases.
A Windows-First RTX Spark Laptop
The RTX Spark laptop platform appears in machines including the MSI Prestige, Asus ProArt and Dell XPS, according to MindStudio. The report says these systems use the same GPU design as Nvidia’s DGX Spark desktop but ship with Windows on ARM, while the desktop unit comes with Nvidia’s Linux build. That difference makes laptop Linux support especially relevant to developers accustomed to Linux-first AI software.
Getting Linux working on the tested MSI system was not plug-and-play. MindStudio describes installation attempts that encountered display problems, missing keyboard, touchpad and GPU support, and a need for a USB Ethernet adapter. Using a kernel with Nvidia’s open-source driver eventually enabled a Linux desktop, CUDA acceleration and 4K output at 120Hz, after further configuration. These setup details show that the reported performance gap came after the system had been made usable, rather than from an inability to boot Linux at all.
The report’s benchmark results also varied by operating environment: WSL is a Linux environment running within Windows, not a native Linux installation. MindStudio says it reached 65.1 tokens per second on GPT-OSS-120B, close to Windows’ 65.4. That comparison is relevant because WSL continues to use the Windows driver stack, which the report says has access to the platform power controls.
““Windows produced 65.4 tokens per second. WSL, which runs a Linux environment inside Windows, hit 65.1 tokens per second. Native Linux on the laptop, however, dropped to 13.9 tokens per second.””
— MindStudio
Nvidia Response and Wider Tests
The supplied material does not include a response from Nvidia or confirmation that the company plans to add equivalent power-management support to its Linux driver. It also does not provide the report’s publication date, full testing methodology, driver and kernel versions, or enough detail to assess how the results might change across other RTX Spark laptop models and Linux distributions.
The article’s excerpt ends while discussing performance on battery power, so the reported plugged-in results do not establish how native Linux performs unplugged. It is also unclear whether firmware or driver updates released after the tests have changed the clock behavior. The memory-allocation workaround is described as requiring Windows, but the source does not establish whether another supported configuration method exists.
Driver Support Is the Key Test
The immediate technical question is whether Nvidia will provide Linux with access to the platform’s power-management logic or otherwise allow the GPU to reach its intended clocks. Until a driver or firmware change is documented and tested, the report’s findings point to WSL as the practical workaround for users who want Linux tools on the tested laptop without giving up Windows-managed GPU performance.
Further reporting would need to compare updated drivers across multiple RTX Spark laptop models, Linux distributions and power conditions, while publishing the benchmark setup and repeatability data. Buyers considering native Linux should check for model-specific support and independent tests rather than assuming that CUDA availability alone guarantees full-speed inference.
Key Questions
How much slower was native Linux in the reported test?
MindStudio measured 13.9 tokens per second on native Linux for GPT-OSS-120B, versus 65.4 on Windows on the tested MSI Prestige RTX Spark laptop.
What does the report say causes the slowdown?
It attributes the low GPU clocks to a Windows-only Nvidia platform power-management component and encrypted firmware tables that Linux cannot currently read. Nvidia has not confirmed that explanation in the supplied material.
Does WSL have the same performance problem?
Not in MindStudio’s reported test. WSL reached 65.1 tokens per second, close to Windows’ 65.4, apparently because it uses the Windows driver stack.
Can Linux run on the tested laptop at all?
Yes, according to MindStudio, after installation and hardware configuration work. The report says it achieved a Linux desktop, CUDA acceleration and 4K output at 120Hz, but with much lower plugged-in inference speed.
Has Nvidia said it will fix the issue?
The supplied source contains no Nvidia response or announced fix. MindStudio says a remedy would require Linux driver support for the platform’s power-management logic.
Source: mindstudio.ai
