# Standard benchmark runs

Strata's `standard-v1` suite (`strata.py bench run`, protocol `strata-qwen38-thinking-v1-draft.1`): 3×
short / 4K / 16K prompts with a 512-token cap at the Qwen model-card sampling (temperature 1.0, thinking
on, effort xhigh), then 12 sanity cases (instruction, retrieval, tool calls). Each folder is the runner's
own output: `README.md` (its report), `results.json`, `config.json`, `outputs.jsonl`, `provenance/`.

Engine 0.1.51 at `4b9e05a2`, gpu-unlock, idle server, run from the guest over loopback. Tuning arms and
the power-cap runs use IQ3_S ([`strata-iq3_s.json`](../../strata/strata-iq3_s.json)); the quant comparison
at the end says which config each run used.

## Tuning arms

Each arm = the setup config ([`strata/strata-iq3_s.json`](../../strata/strata-iq3_s.json)) plus one
change, a fresh engine restart, then the suite. Median of 3; decode and prompt are the engine's timings.

| Arm | Folder | Prompt 4K tok/s | Prompt 16K tok/s | TTFT 16K | Decode short | Decode 4K | Decode 16K |
|---|---|---:|---:|---:|---:|---:|---:|
| Warm baseline (15 h uptime, + concurrency stage) | `0f07e24d` | 1,776 | 2,938 | 5.61 s | 130.7 | 123.2 | 124.5 |
| **Restart baseline** | `87dfe6da` | **1,712** | **2,881** | **5.71 s** | **122.5** | **110.1** | **116.6** |
| `--adapt-async 1` | `1ecae1bc` | 1,712 | 2,870 | 5.73 s | 125.1 | 115.0 | 108.6 |
| `--pipeline-windows 2` | `5f934909` | 1,705 | 2,878 | 5.71 s | 110.0 | 104.2 | 102.7 |
| `--prefill auto:32768` | `92972582` | 1,702 | **2,239** | **7.35 s** | 122.7 | 111.6 | 105.9 |
| all three | `53bbc5d8` | 1,699 | 2,243 | 7.34 s | 121.1 | 121.7 | 113.4 |

All 12 sanity checks passed in every arm. The arms report "nonconforming" only because they used
`--skip-model-hash` and `--track custom`. GPU temps peaked at 56 / 64°C (GPU1 ~230 W).

Four requests at once (warm baseline, server capacity 1): 115.5 / 126.5 / 126.2 tok/s total per round,
so concurrent clients just queue.

## Conclusions

- **Keep the setup defaults.** None of the three flags helped on this machine.
- `--prefill auto:32768` is clearly worse at 16K: −22% prompt speed, +1.6 s to first token. Strata's
  docs say chunks above 8192 aren't bit-exact on CUDA anyway. The gains reported elsewhere were at 74K
  prompts on AMD; very long prompts were not tested here.
- `--pipeline-windows 2` left prompt speed unchanged and cost decode speed (−10% short, −12% at 16K). The
  community 2×3090 report's +17% prefill did not reproduce (their cards were x16 + x4; ours are x8 + x8).
- `--adapt-async 1` is within noise.
- **Noise:** decode varies 94–135 tok/s *within* an arm (sampling at temperature 1.0 changes how many
  draft tokens are accepted), so decode differences under ~10% between single runs aren't meaningful.
  Prompt speed is tight (±2%) and is the reliable comparison.
- Prompt speed is already above the community 2×3090 report (1,798 tok/s at 19.9K). The 915 tok/s in
  the earlier notes was a first cold prompt right after load, not a steady-state number.

## Power cap: 250 W vs 275 W

2026-10-27, setup defaults, **case closed**, caps set with `nvidia-smi -pl` at runtime. Runs alternated
250 / 275 / 250 / 275, each after a fresh restart. GPU temp, power, SM clock and throttle reasons were
sampled every 2 s.

| Cap | Folder | Prompt 4K tok/s | Prompt 16K tok/s | TTFT 16K | Decode short | Decode 4K | Decode 16K |
|---|---|---:|---:|---:|---:|---:|---:|
| 250 W | `6a05efdc` | 1,719 | 2,897 | 5.69 s | 122.9 | 112.9 | 108.2 |
| 275 W | `5c64d992` | 1,742 | 3,030 | 5.44 s | 125.3 | 115.3 | 116.1 |
| 250 W | `ae1396b5` | 1,712 | 2,887 | 5.71 s | 116.3 | 122.1 | 106.8 |
| 275 W | `16ea71d0` | 1,743 | 3,016 | 5.46 s | 114.2 | 111.6 | 111.0 |

While busy (samples above 150 W):

| Cap | GPU | Peak temp | Mean / peak W | Mean SM MHz | Power-throttled |
|---|---|---:|---:|---:|---:|
| 250 W | 0 | 51°C | 220 / 250 | 1,858 | 62 of 80 samples |
| 250 W | 1 | 59°C | 223 / 249 | 1,807 | 82 of 86 |
| 275 W | 0 | 53°C | 228 / 273 | 1,867 | 45 of 78 |
| 275 W | 1 | 63°C | 239 / 272 | 1,839 | 80 of 87 |

- **275 W: +4.6% prompt speed at 16K** (2,892 → 3,023 tok/s, −0.25 s to first token), +1.6% at 4K. Both
  repeats agree to within 0.5%, so this is real and not noise.
- Decode is unchanged within noise; decode isn't power-bound.
- Cost: +2–4°C peak and +8–16 W mean per card while busy. All 12 sanity checks passed in every run.
- GPU1 still hits its power limit most of the time at 275 W, so a higher cap would probably buy a
  little more prompt speed.

## Quant: IQ3_S vs UD-Q4_K_XL

2026-10-27, 275 W, case closed, setup defaults for each quant
([`strata-iq3_s.json`](../../strata/strata-iq3_s.json),
[`strata-unsloth-ud-q4_k_xl.json`](../../strata/strata-unsloth-ud-q4_k_xl.json)), fresh restart before each
run. IQ3_S rows are the two 275 W runs from the power-cap test.

| Quant | Folder | Prompt 4K tok/s | Prompt 16K tok/s | TTFT 16K | Decode short | Decode 4K | Decode 16K |
|---|---|---:|---:|---:|---:|---:|---:|
| IQ3_S | `5c64d992` | 1,742 | 3,030 | 5.44 s | 125.3 | 115.3 | 116.1 |
| IQ3_S | `16ea71d0` | 1,743 | 3,016 | 5.46 s | 114.2 | 111.6 | 111.0 |
| UD-Q4_K_XL | `58540dcc` | 846 | 1,918 | 8.57 s | 66.9 | 69.7 | 67.8 |
| UD-Q4_K_XL | `3eb539d8` | 847 | 1,918 | 8.57 s | 65.7 | 68.9 | 76.3 |

- **UD-Q4_K_XL is much slower:** prompt −51% at 4K and −37% at 16K (+3.1 s to first token at 16K),
  decode −40% (~66–76 vs ~111–125 tok/s). Both repeats agree to within 0.2% on prompt speed.
- Why: its experts are 71.7 GiB vs 47 GiB, so only 5,328 fit in VRAM (15.6 GiB) vs 9,508 for IQ3_S, and
  every miss moves more bytes. Load took 2 min 22 s; 82 GB of the guest's RAM in use, 27 GB available.
- The GPUs work less hard: mean 211–227 W while busy (IQ3_S 228–239 W), peaks 54 / 63°C.
- All 12 sanity checks passed in both runs. Quality differences these checks can't see are left to
  husky-bench; Strata's yardstick puts UD-Q4_K_XL at 93.1% same-top-token as Q8_0 vs 87.7% for IQ3_S.

## Not tried yet

- Huge pages for the expert arena: the engine logs `MAP_HUGETLB unavailable (needed 23983 2 MiB pages,
  vm.nr_hugepages=0)`. That would mean reserving ~47 GB of huge pages in the guest (a NixOS change and a
  reboot).
- Power cap above 275 W (300–350 W). gpu-unlock GEMM was +40% at 350 W, but 275 W gave only +4.6% prompt speed.
- Other quants: UD-IQ4_XS (barely better than IQ3_S on Strata's yardstick), IQ3_XXS / IQ2_XS (smaller,
  faster, worse).