‹ BackHN Continuity

Thread

Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

741 points · 330 comments · snehesht

  1. quietFalcon · · focus · HN ↗
    Nice, though generation speed is the easy half for MoE offload, what's your prompt processing look like at say 16k context?
    1. snehesht · · focus · HN ↗
      They have some community benchmarks published <a href="https:&#x2F;&#x2F;github.com&#x2F;Niko1221&#x2F;Strata&#x2F;tree&#x2F;main&#x2F;bench&#x2F;results" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;Niko1221&#x2F;Strata&#x2F;tree&#x2F;main&#x2F;bench&#x2F;results
    2. merbanan · · focus · HN ↗
      Q2_0 does 33 tok&#x2F;s decode and ~600t&#x2F;s prompt processing at 128k context on RTX2060 8GB VRAM.

      ISTA IQ3_XXS does ~21 tok&#x2F;s decode and ~240t&#x2F;s prompt processing

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.