Running Arch Linux on a 128GB Laptop for Local AI Models
Original source
Putting Linux on my RTX Spark laptop because I don't have a Microsoft account
Hacker News →A developer installed Arch Linux ARM on an ASUS ProArt P16 with 128 GB RAM and 6,144 CUDA cores, avoiding Windows due to privacy, advertising, and interface design concerns. This required custom kernel patches and hardware packages to support the NVIDIA GPU driver and peripherals like the keyboard and audio subsystem.
CPU performance matched Windows closely. In 7-Zip benchmarks, the Linux setup performed about 5% faster on compression and 3% faster on decompression than comparable Windows configurations. The more significant advantage lies in memory architecture: Linux presents the full 116.0 GiB as a single pool available to both CPU and GPU workloads, whereas Windows partitions RAM into dedicated GPU, shared, and CPU-only regions, limiting resource allocation flexibility.
Testing with the Qwen3.8-Flash-Next language model demonstrated the system’s suitability for local inference. The system allocated nearly 90 GB for the model and key-value cache, achieving roughly 30 tokens per second on code review tasks and 86.6 tokens per second when handling four concurrent requests. Stability testing over ten minutes with 150 requests showed consistent performance with all responses passing correctness validation.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.