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 measured GPT-OSS-120B inference at 13.9 tokens per second on native Linux on an MSI Prestige RTX Spark laptop, compared with 65.4 on Windows and 65.1 in WSL. The report attributes the gap to Linux’s inability to access encrypted power-management firmware that lets the Windows driver raise GPU clocks; Nvidia has not provided a response in the supplied material.
A MindStudio report found that native Linux inference on an MSI Prestige RTX Spark laptop ran at 13.9 tokens per second in a GPT-OSS-120B test, compared with 65.4 on Windows and 65.1 in Windows Subsystem for Linux. The report attributes the roughly fivefold performance gap to a Linux power-management limitation: the GPU stayed near 780 MHz because Linux could not access a firmware power table available through the Windows driver.
MindStudio said the laptop ran Linux with a desktop environment, CUDA acceleration and functioning audio, keyboard and touchpad after configuration. But the GPU’s clock speed was reported at about 780 MHz on Linux, compared with roughly 2.2 GHz on Windows while plugged into wall power. In the GPT-OSS-120B benchmark, native Linux produced 13.9 tokens per second; Windows reached 65.4, WSL reached 65.1 and Nvidia’s DGX Spark desktop reached 61.
A separate test using vLLM with eight concurrent clients also showed lower throughput on native Linux: 39.5 tokens per second, versus 127 in WSL, according to MindStudio. The report said a longer run took about 37 minutes on Linux and eight minutes on Windows. Temperatures were reported between 40 and 47 degrees Celsius during the Linux run, which MindStudio said pointed away from thermal throttling and toward the GPU not being allowed to reach higher clocks.
Installation and memory configuration presented separate hurdles. Linux initially saw only 28GB of a laptop configured with 128GB of unified memory because the firmware reserved 96GB for GPU use. MindStudio said changing that allocation in Windows to reserve 4GB for the GPU allowed Linux to see 120GB. The report describes this setting as unavailable from Linux.
Linux AI Workloads Lose GPU Speed
The results matter to developers considering these laptops for local AI inference. A system advertised with substantial unified memory and an Nvidia GPU may run Linux-first tools, but the reported performance differs sharply depending on whether workloads use native Linux or the Windows driver stack. For the tested model, WSL delivered throughput close to Windows, while native Linux was substantially slower.
That creates a practical trade-off rather than a simple compatibility verdict. Developers can use Linux applications through WSL and retain near-Windows performance in the reported tests, but those who need a native Linux installation may face slower inference and additional setup work. The tests are from one report and one laptop configuration, so they do not establish the same performance gap across every RTX Spark laptop, model or workload.
Products Worth Considering
Samsung Galaxy Tab A11+ (6 GB RAM, 128 GB storage, expandable) – Gray
Apple MacBook Air 13.6-inch with M5 Chip, 16GB Memory, 512GB SSD, Sky Blue
Apple MacBook Air 15-inch with M5 chip, 24GB RAM, 512GB SSD, Midnight
As an affiliate, we earn on qualifying purchases.
The Windows Driver Difference
The RTX Spark laptop platform appears in machines from manufacturers including MSI, Asus and Dell, according to the source. MindStudio describes the laptops as using the same GPU design as Nvidia’s DGX Spark desktop, but says the laptop versions ship with Windows on ARM, while the DGX Spark ships with Nvidia’s Linux system. The laptops’ unified memory, available in configurations up to 128GB, is aimed in part at running AI workloads locally.
MindStudio attributes the clock-speed difference to the Nvidia Platform Controllers and Framework, a Windows-only component that manages power delivery between the CPU and GPU. The report says the power table is encrypted and Linux cannot read it, while the Windows driver can. It also describes Linux installation as requiring hardware-specific workarounds, including a USB Ethernet adapter for installation and changing the memory split in Windows before booting Linux.
“Windows produced 65.4 tokens per second. WSL, which runs a Linux environment inside Windows, hit 65.1 tokens per second.”
— MindStudio report
Limits of the Benchmark Findings
The supplied report does not establish whether the same clock limits affect every RTX Spark laptop model or Linux distribution. It also does not provide independent replication, full benchmark methodology, or a response from Nvidia. The reported encrypted firmware explanation is MindStudio’s account; details of the firmware interface and whether Nvidia plans to change Linux support are not included in the supplied material.
The source text begins to describe a different result when the laptop is unplugged but cuts off before giving that comparison. The behavior on battery power therefore cannot be summarized reliably from the material provided. It is also unclear whether later driver, kernel or firmware updates could alter native Linux performance.
Nvidia Support Is the Key Fix
MindStudio says a lasting fix would require Nvidia to bring the platform controller’s power-management logic to its Linux driver. The supplied material does not say Nvidia has announced such work or set a release date. Until a change is documented and tested, users weighing the laptop for Linux AI workloads can compare their needs against the report’s findings and consider WSL, which retained near-Windows performance in the cited tests.
Further testing across other RTX Spark models, Linux distributions, power conditions and inference workloads would help determine how broadly the issue applies. Nvidia’s response, if issued, and any driver or firmware updates would clarify whether native Linux can access the same GPU boost behavior as Windows.
Key Questions
How fast was native Linux in the reported test?
MindStudio measured 13.9 tokens per second for GPT-OSS-120B on native Linux on an MSI Prestige RTX Spark laptop. The report measured 65.4 on Windows and 65.1 in WSL.
What does the report say causes the slowdown?
MindStudio attributes it to power management. It says Linux cannot read an encrypted firmware power table used by a Windows-only Nvidia platform controller, leaving the GPU near 780 MHz in the tested plugged-in configuration.
Does Linux run on the RTX Spark laptop?
Yes, according to the report, after configuration and hardware workarounds. MindStudio said it achieved a working Linux desktop with CUDA acceleration and functioning audio, keyboard and touchpad.
Is WSL a faster option for Linux AI tools?
In the cited tests, yes. WSL reached 65.1 tokens per second on GPT-OSS-120B and 127 tokens per second in a vLLM test with eight concurrent clients. Those are results from MindStudio’s testing, not a guarantee for all workloads.
Has Nvidia confirmed a fix?
The supplied material includes no Nvidia response or announced fix. MindStudio says a remedy would require Nvidia to add the relevant power-management logic to its Linux driver.
Source: mindstudio.ai
