Agent step time on the Genesis box: 12.4 s median before, 484 s after (a 39x slowdown)
Hardware and Config
- GPU: 2x NVIDIA RTX PRO 6000 Blackwell Max-Q 96 GB
- CPU: AMD Threadripper PRO 9975WX
- OS: Windows 11
- Date: see the source CSV timestamps
- What was measured: OpenCode assistant step durations (model generation time, not tool time)
Software Versions
- Model: genesis-qwen (the Genesis LLM on FORGE)
- Client: OpenCode (DAD-PC, Windows 11)
- Date: see the source CSV timestamps (pre-slowdown and post-slowdown windows)
Raw Numbers
Two windows of OpenCode assistant steps, model_seconds (step start to streamed completion):| Window | Steps | Model median (s) | Tool median (s) |
|---|---|---|---|
| Pre-slowdown (17:00 to 18:12 ET) | 143 | 12.4 | 0.0 |
| Post-slowdown (18:12 ET onward) | 80 | 484.0 | 0.0 |
What the Numbers Do and Do Not Show
- DO show: the median model-generation time per OpenCode step, before and after the slowdown event on Nov 5, with the tool time separated (and confirmed negligible)
- DO NOT show: the cause of the slowdown (still under investigation by the Genesis operator), the per-step distribution (median only), or whether the slowdown has since recovered. A 39x median jump from 12.4 s to 484 s is a real operational event, not noise: 143 pre-slowdown steps and 80 post-slowdown steps, both with 0 s tool median.
AI Disclosure
This benchmark was run and written by an AI agent (OpenCode Build for Danny) on the Genesis box. All numbers are from the attached raw-numbers file. The AI disclosure line is required on every post.Link Slots
- Raw numbers: docs/recipes/data/STEPTIME-001.json
- Source CSV: docs/recipes/raw/pre-slowdown-step-times-20261105.csv
- Related posts: IMGGEN-001 (image gen), NETLAT-001 (network latency)
Status
READYWindow dates (for the record)
- Pre-slowdown: 17:00 to 18:12 ET on the afternoon of 5 Nov
- Post-slowdown: 18:12 ET onward
- The source CSV carries the full timestamp; the numbers above are medians only.