#3155 Community benchmark on A6000 (Turing)

Description

# Community benchmark on NVIDIA RTX A6000

Measured on 2026-10-19 with `tools/strata_bench.py` (suite 1, 3 runs per test, medians with ranges). Speeds are the engine's own timings; first-token times are measured by the client over HTTP. Speed only: nocorrectness check was run.

## Hardware and software

- GPU and VRAM: NVIDIA RTX A6000, 45 GiB VRAM
- CPU: AMD Ryzen 9 5950X 16-Core Processor (16 cores, 32 threads); installed RAM: 62.7 GiB; storage: not measured
- OS: Linux 7.2.11-gentoo; GPU driver: 615.78.08; CUDA: not read
- Strata: project 0.1.45, commit 03edd7ce; engine 0.1.45
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: qwen3.8-flash-next-coder-iq1_m
- Context: 262144 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking on for the suite; temperature 0

```text
strata --pack coder-iq1_m --native Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ1_M-00002-of-00002.gguf --expert-profile expert-profile-coder.bin --expert-cacheauto --prefill auto --spec 4 --mtp rt --max-context 262144 --kv int8 --kv-resident 32768 --vram-reserve-mib 64 --spec-min-p 0.5
```

## Method

Run by `python tools/strata_bench.py --long --think` against the HTTP API of a Strata server (already running); model loading is not timed. One warm-up chat and a 1K prompt are discarded first. Every prompt starts with a unique tag, so the conversation cache cannot help the fresh-prompt tests.

- short chat: the three questions of `tools/ab_engine.py`, up to 100 new tokens, 3 runs
- 4K / 16K prompts: generated word lists sized by a calibration request, 24 new tokens, 3 fresh prompts each
- agent session: system prompt and four tools, then 5 turns that each add about 1500 words of a project file and continue from the model's own earlier answers (the way agent clients resend a conversation), 64new tokens per turn, 3 sessions; the cache numbers come from the answer's usage (`cached_tokens`) and timings (`prompt_n`)
- concurrency (when run): N requests sent at the same moment, total tokens over the wall time of the burst, latency over all requests

## Results

| Configuration | Actual prompt tokens | Reused tokens | Generated tokens | Runs | Prompt tok/s median and range | Decode tok/s median and range | TTFT seconds median and range |
| --- | ---: | ---: | ---: | ---: | --- | --- | --- |
| short chat | 86 (83-91) | 0 | 100 (100-100) | 9 | not measured | 95.4 (91.2-112.1) | 0.34 (0.34-0.34) |
| 4K prompt, fresh | 4059 (4059-4059) | 0 (0-0) | 24 cap | 3 | 2012.3 (2006.6-2016.1) | not measured | 2.04 (2.03-2.04) |
| 16K prompt, fresh | 16098 (16098-16098) | 0 (0-0) | 24 cap | 3 | 2602.3 (2600.2-2609.5) | not measured | 6.21 (6.20-6.22) |
| 32K prompt, fresh | 32089 (32089-32089) | 0 (0-0) | 24 cap | 3 | 2689.4 (2668.2-2690.3) | not measured | 11.97 (11.96-12.06) |
| agent session, follow-up turns | 14504 (14504-14524) at the last turn | see below | up to 64 | 3 | see below | 101.7 (92.9-110.0) | 1.75 (1.19-1.93) |

Ranges are the lowest and highest of the runs; `Runs` counts requests that returned a token.

### Agent session (conversation cache)

Cache hit over the follow-up turns: 68.3 % (68.3-68.3); earlier-turn tokens read again per session: 345 (345-345) (earlier turns that the cache did not hold; the 13539 (13539-13559) new tokens of the files are read once); first turn (nothing cached) first token: 0.62 s (0.62-0.64).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 645 | 0 | 645 | 1043.7 | 0.636 | 96.8 |
| 1 | 3790 | 640 | 3150 | 1740.8 | 1.833 | 96.0 |
| 2 | 5280 | 3785 | 1495 | 1272.2 | 1.196 | 110.0 |
| 3 | 8681 | 5275 | 3406 | 1784.6 | 1.932 | 99.2 |
| 4 | 11562 | 8676 | 2886 | 1677.1 | 1.744 | 107.1 |
| 5 | 14504 | 11557 | 2947 | 1703.0 | 1.753 | 109.9 |

Per-request data: `results.json`. Engine log lines from the start: `engine-log-excerpt.txt` (when this run started the server).

## Correctness and limitations

Not measured: this is a speed suite. The conversation uses the model's own replies, so token counts differ a little between runs. Machine details come from the server, `nvidia-smi`/`rocm-smi` and the OS; background load and power limits are not recorded. Total time of the suite: 116 s.