Auto-fit vs tuned MoE offload: 564 → 1330 pp tok/s, unchanged decode (Qwen3.6-35B-A3B Q6 / RTX 3090)
英文标题：Auto-fit vs tuned MoE offload: 564 → 1330 pp tok/s, unchanged decode (Qwen3.6-35B-A3B Q6 / RTX 3090)
TL;DR: On a Qwen3.6-35B-A3B Q6 setup sized for 64K context on a 24GB RTX 3090, spilling eight MoE expert layers to CPU freed enough VRAM to increase -b from 512 to 1024 and -ub from 128 to 512. Prompt processing improved by 2.36×, while generation speed remained unchanged within measurement noise.
打开原文相关 2 · 机会 0
中文
TL;DR: On a Qwen3.6-35B-A3B Q6 setup sized for 64K context on a 24GB RTX 3090, spilling eight MoE expert layers to CPU freed enough VRAM to increase -b from 512 to 1024 and -ub from 128 to 512. Prompt processing improved by 2.36×, while generation speed remained unchanged within measurement noise.
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
TL;DR: On a Qwen3.6-35B-A3B Q6 setup sized for 64K context on a 24GB RTX 3090, spilling eight MoE expert layers to CPU freed enough VRAM to increase -b from 512 to 1024 and -ub from 128 to 512. Prompt processing improved by 2.36×, while generation speed remained unchanged within measurement noise.