the lab

every page is a test i ran at home: the setup, the numbers, what went wrong

  1. Oct 20, 2026local ai

    two coding agents on the same box, from 44 to 66 tok/s each in one evening

    43.8 → 66.4 tok/s per agent

    same machine, same model, one night of tuning: when two agents generate at the same time, each one goes from 43.8 to 66.4 tok/s, and four settings did most of the work

  2. Oct 20, 2026local ai

    glm 5.3 flash (321B) on a 3090 + a 3080 ti: from 26 to 31.7 tok/s, and why i'm still sticking with qwen

    26.1 → 31.7 tok/s

    two maya options give 22% more decode to a 321B model that spills way over into ram, but on a real agent task it stays clearly slower than qwen 3.8 flash next on the same box

  3. Oct 20, 2026local ai

    a 4B model holds 256K of context on a single 12gb 3080 ti, and finds 40 codes out of 40

    136 → 35 tok/s at 254K

    spark-x2.6-4b in q8 with llama.cpp: 136 tok/s empty, still 35 tok/s with 254K tokens of context, and a 12gb card that beats the 3090 because i had capped it less

  4. Oct 20, 2026hardware & power

    i undervolted my 3090 and my 3080 ti on linux, -94 W for almost nothing

    452 W → 358 W

    two coding agents working at the same time, 452 W → 358 W across both cards, the fans go from 91% to 48%, and throughput loses 7%

  5. Oct 19, 2026local ai

    i fit 400K of context per agent on 80gb of ram, and it still finds a code hidden at 92% of the prompt

    3/3 codes at 373,723 tokens

    two slots of 409,600 tokens fit on 80gb of ram, and a 373,723 token prompt with three codes hidden inside gets 3 right answers out of 3, read in 195 seconds

  6. Oct 19, 2026crashes

    i asked for 512K of context × 4 slots on 80gb of ram, and the whole machine froze

    ~23gb of cache too many

    at first i blamed a dead wifi dongle, but it was the ram: the session cache at 512K needed ~23gb more on top of a 40gb model that was already taking up the space, and a simple memory cap would have avoided it all

  7. Oct 19, 2026local ai

    strata 0.2.0 on my rtx 30 cards: +12 to +17% decode, but my agents don't finish any faster

    91.7 → 107.6 tok/s

    a simple engine update takes solo decode from 91.7 to 107.6 tok/s on a 3090 + 3080 Ti, and yet the total time of my agents doesn't move, because they spend their life reading prompts