#2424 Community Benchmark Report: RTX A6000 48GB / Qwen3.8-Flash-Next IQ3_S

Description

# Community benchmark on NVIDIA RTX A6000

Measured on 2026-10-14 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: no correctness check was run.

## Hardware and software

- GPU and VRAM: NVIDIA RTX A6000, 48 GiB VRAM
- CPU: Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz (48 cores, 48 threads); installed RAM: 125.4 GiB; storage: not measured
- OS: Linux 5.14.0-611.55.1.el9_7.x86_64; GPU driver: 580.159.03; CUDA: 13.0
- Strata: project 0.1.44.1, commit not a git checkout; engine 0.1.44.1
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: qwen3.8-flash-next-iq3_s
- Context: 131072 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata --pack iq3_s --native Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --mtp rt --max-context 131072 --kv int8 --kv-resident 32768 --spec-min-p 0.5 --pool-workers 24
```

## Method

Run by `python tools/strata_bench.py` against the HTTP API of a Strata server (started by the benchmark from the install configuration); 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), 64 new 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 | 46 (43-51) | 0 | 100 (100-100) | 9 | not measured | 108.6 (97.5-146.9) | 0.30 (0.26-0.32) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 2183.9 (1877.0-2190.5) | not measured | 1.86 (1.86-2.16) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 2799.9 (2711.4-2807.8) | not measured | 5.77 (5.76-5.96) |
| agent session, follow-up turns | 14535 (14532-14557) at the last turn | see below | up to 64 | 3 | see below | 106.5 (69.0-148.5) | 1.79 (1.12-2.36) |

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: 68.7 % (68.6-68.8); earlier-turn tokens read again per session: 4 (3-74) (earlier turns that the cache did not hold; the 13729 (13729-13729) new tokens of the files are read once); first turn (nothing cached) first token: 0.79 s (0.71-0.84).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 749.2 | 0.836 | 84.6 |
| 1 | 3752 | 631 | 3121 | 1336.3 | 2.363 | 106.5 |
| 2 | 5285 | 3815 | 1470 | 1088.8 | 1.379 | 99.8 |
| 3 | 8729 | 5278 | 3451 | 1486.3 | 2.362 | 88.8 |
| 4 | 11614 | 8752 | 2862 | 1346.8 | 2.17 | 107.5 |
| 5 | 14532 | 11633 | 2899 | 1442.6 | 2.046 | 107.2 |

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: 74 s.