# Putting Linux on My RTX Spark Laptop

> My Arch Linux setup on the ASUS ProArt P16 beats Windows in 7-Zip tests and reports 116.0 GiB memory for CUDA.
- Title: Putting Linux on My RTX Spark Laptop · Josie | Shortify
- Summary: My Arch Linux setup on the ASUS ProArt P16 beats Windows in 7-Zip tests and reports 116.0 GiB memory for CUDA. Josie's Arch Linux on ASUS ProArt P16 runs…
- Keywords: arch linux, asus proart p16, nvidia rtx, cuda, 7-zip, memory pool, linux distribution, open-weight models, qwen, vllm, technology, startups
- Source: Josie — https://josie.lol/blog/linux-rtx-spark
- Author: Josephine Pfeiffer
- Published: 2026-10-19T00:00:00+00:00
- Read time: 1 min
- Topics: arch linux, asus proart p16, nvidia rtx, cuda, 7-zip, memory pool, linux distribution, open-weight models, qwen, vllm, technology, startups
## Why I switched to Linux
I wanted to run open-weight models locally on my high-end laptop without Microsoft's privacy issues.
## My laptop specs
ASUS ProArt P16 H7607BA-SE033W with 128 gb ram, 6144 cuda cores, and a 2 tb ssd.
## Why Windows is problematic
Microsoft's setup requires privacy concessions and takes years to fix basic features like the taskbar.
## My Linux setup
I used Arch Linux ARM with patches for NVIDIA and ASUS hardware to create a bootable system.
## 7-Zip performance
My Linux setup compresses 20 threads 5% faster than Windows and decompresses 3% faster.
## Memory pool comparison
Linux reports 116.0 gi b as both system ram and cuda memory, while Windows divides ram into dedicated and shared regions.
## Running Qwen locally
I ran NVIDIA's Qwen3.8-Flash-Next-FP4-Local checkpoint through vLLM with mixed quantization.
## Model performance
Qwen processed 150 requests in ten minutes with all answers passing checks and median response times under 6 seconds.
## Why this matters
This setup allows running complex AI models locally on high-end laptops without microsoft dependencies.
## Key takeaway

Josie's Arch Linux on ASUS ProArt P16 runs 7-Zip 20-thread compression 5% faster than Windows and reports 116.0 GiB memory pool for CUDA.