Buzzing HN · 中文精选· novaRom·· 7 hours agoSelectedAI score71
Benchmarking Why RTX Spark Laptops Drop to 35% Inference Speed When Plugged In on Linux
在 Linux 系统上，Nvidia RTX Spark 加密固件将推理速度降至 35%
AI guide
When running native Linux on RTX Spark-based laptops (such as the MSI Prestige N16 Flip AI Plus) while connected to AC power, GPU clock speeds are locked at around 780 MHz. This causes GPT-OSS-120B inference speeds to plummet from 65.4 tokens/s on Windows down to 13.9 tokens/s.
Why it matters
The article benchmarks the root cause—power management firmware throttling compute performance under native Linux—and offers WSL as a viable workaround for developers running local LLMs.
Source:Buzzing HN · 中文精选 · mindstudio.ai