Unverified60% confidenceFactTime unknown
千问3.6 27B FP8模型开启MTP后正常速度约38 token/s，平均测速可达70 token/s以上
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
5/31/2026
First Seen
Valid until: 8/29/2026
Sources
4×3080Ti本地部署千问3.6 27B跑OpenCode编程实测
bilibili苏不二师兄
Related Entities
Related Claims
UnverifiedRTX 5090上Qwen3.6 27B模型Q4量化不开MTP解码速度约63 token/s，开启MTP后跳到105 token/s78% similarUnverifiedQwen3.6 35B模型借助MTP技术在oMLX上达到了86.7 tokens/s的生成速度，prompt处理速度为1735 tokens/s72% similarUnverifiedRTX 5090上Qwen3.6 27B模型预填充速度从174提升到253 token/s，显存占用约18GB71% similarUnverified在显存充足、模型完整驻留GPU时，RTX 4090跑Q4量化的27B模型通常可达15～30 tokens/s；触发CPU卸载后速度可能骤降至1～2 tokens/s甚至更低70% similarVerified在RTX 5090上，Muse Glimmer生成速度从每秒75个Token提升到233个Token，提速约3.1倍70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/4397API
curl https://kongchang.com/api/v1/knowledge/claims/4397MCP
get_claim(id=4397)