MiniMax-M3.1-Flash at 3.5 bpw needs about 12 GiB more weights per Spark than 3.06 bpw…
This is a dev post classified by AGK Intelligence as Other (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
MiniMax-M3.1-Flash at 3.5 bpw needs about 12 GiB more weights per Spark than 3.06 bpw and decodes at 23.6 vs 23.9 tok/s 3.06 bpw, 2 DGX Sparks, 256 GB Decode, code: 24.7 tok/s Decode, prose: 23.0 tok/s 500 words (code and prose): about 28.2 s First token at 8K: 17.7 s Peak memory on the busier Spark: 93.9 GiB 3.5 bpw, 2 DGX Sparks, 256 GB Decode, code: 24.7 tok/s Decode, prose: 22.5 tok/s 500 words (code and prose): about 28.0 s First token at 8K: 17.8 s Peak memory on the busier Spark: 107.9 GiB Mia's EXL3 EAQ quants, DSpark drafts, 32,768 window, 2 slots, 3 GiB prompt cache Speed test, me
Posted by Niklas (2.8k followers) 1 h ago · 2 likes · 187 views · view the original post on X. Kept by the Dev Radar as Other.
More dev work like this
- Wait, Seedance 2.5 at roughly half Higgsfield’s API price? — @wallstengine
- X just pushed more For You Recommendation Algorithm code to GitHub. Credit to… — @blankspeaker
- Testing Maya's Turbo mode 🔥 — @PeasantSmith
- David explain exactly how 'Abliterated uncensored' works broke down the technique and… — @0x0SojalSec
- AI agents can now trade on Cancore – within limits you control. — @cancore_io
- AI agents can now trade on Canton. — @CantonNetwork
- Run the NEW GLM 5.3 Flash recipe (1104) on your 2x NVIDIA DGX Sparks with ease ✨ — @MiaAI_lab
- 震惊！小米人形机器人在自家汽车工厂连续自主干了8小时05分钟，还自己纠错、和产线配合分拣零件。 — @NFT_Chen
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.3k posts from 5k X accounts over the last 21 days, 9.3k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-04 19:07 UTC. Full method.