a contributor put two used rtx 3060 12gb cards on one board, and ran qwen 3.8 27b dense…
This is a dev post classified by Jev as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
a contributor put two used rtx 3060 12gb cards on one board, and ran qwen 3.8 27b dense at 131k context. 34.2 tok/s stock, 58.9 tok/s with the flag, +72%, on 24gb of vram that costs less than one used 3090 on marketplace. the same pair on the unsloth q4 file at 64k: 30.0 tok/s to 53.3 tok/s, +78%, within a hair of a 3090 power limited to 250w on the same table (52.9 tok/s). the details anon, he ran all 15 arms and kept the losers in. the default layer split leaves 39% on the floor, tensor split first, then the flag. the confidence gate lifts acceptance and costs 21% to 30% of speed here, acc
Posted by Sudo su (36.2k followers) 4 days ago · 165 likes · 11.4k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: qwen38-mtp.
More dev work like this
- A single RTX 3090 hit 381 tok/s on Qwen3.8-27B. — @0x0SojalSec
- AI-written code gets risky when nobody checks the diffs. — @DanKornas
- The new "bakeoff" capability in Compound Engineering is so damn useful. Even on smaller… — @trevin
- Hermes Agent can now stream its THINKING into other AI apps. — @tonysimons_
- Hallmark has been installed 50k+ times! — @nutlope
- Your AI coding agents need a workflow, not more babysitting. — @DanKornas
- Register by October 7 and save >> https://bit.ly/4vkntyD — @linuxfoundation
- Nimble is able to process images! — @madiator
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 13.5k posts from 4.7k X accounts over the last 21 days, 1.5k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-19 19:24 UTC. Full method.