SonicBoom: 52 tok/s decode on a laptop RTX 3050 (llama.cpp: 45) with the same quantized kernels, the difference is a graph compiler. No Python.

1. 

   消息推送

   SonicBoom: 52 tok/s decode on a laptop RTX 3050 (llama.cpp: 45) with the same quantized kernels, the difference is a graph compiler. No Python.

   18:45 · 2026年10月13日 · 周二

   SonicBoom: 52 tok/s decode on a laptop RTX 3050 (llama.cpp: 45) with the same quantized kernels, the difference is a graph compiler. No Python.
   
   🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1x4scub/sonicboom_52_toks_decode_on_a_laptop_rtx_3050/
   🏷️ 分组: Reddit｜AI·自托管·VPS
   🕒 时间: 2026-10-13 17:46:50

   Reddit

   

   From the LocalLLaMA community on Reddit: SonicBoom: 52 tok/s decode on a laptop RTX 3050 (llama.cpp: 45) with the same quantized…

   Explore this post and more from the LocalLLaMA community

   18:45 · 2026年10月13日 · 周二