YOU ARE VIEWING ONE ITEM FROM THE AICRIER FEED

Qwen3.6-Plus throughput drops after idle

AICrier tracks AI developer news across Product Hunt, GitHub, Hacker News, YouTube, X, arXiv, and more. This page keeps the article you opened front and center while giving you a path into the live feed.

// WHAT AICRIER DOES

7+

TRACKED FEEDS

24/7

SCRAPED FEED

Short summaries, external links, screenshots, relevance scoring, tags, and featured picks for AI builders.

Qwen3.6-Plus throughput drops after idle
OPEN LINK ↗
// 56d agoBENCHMARK RESULT

Qwen3.6-Plus throughput drops after idle

A LocalLLaMA user reports 155-160 t/s on a 7900XTX at first boot, then a hard drop to 50 t/s after the machine sits idle for a while. The slowdown persists across context-size changes and only clears after a full PC reboot.

// ANALYSIS

Likely a GPU driver or power-state bug, not a model or context-window problem. The reboot-only recovery pattern is the strongest clue that the runtime or Radeon stack is getting stuck in a bad state.

  • The user says GPU temperature stays around 40C, which makes thermal throttling an unlikely explanation.
  • Context size changes from 32K down to 4K do not fix it, so this does not look like a simple memory-pressure issue.
  • Similar 7900XTX reports in the wild point to idle-power, clock-states, and driver regressions that only reset cleanly after reboot.
  • For local LLM users on AMD hardware, the next variables to isolate are driver version, backend/runtime changes, and any Windows power-management behavior.
// TAGS
qwen3.6-plusllmgpuinferencebenchmark

DISCOVERED

56d ago

2026-04-17

PUBLISHED

56d ago

2026-04-17

RELEVANCE

8/ 10

AUTHOR

soyalemujica