#3169 Community benchmark: 2x RTX 3090 24GB, 192GB RAM, Qwen 3.8 Flash Next IQ3_XXS, 256K Context, Strata 0.1.45

Description

# Community benchmark on NVIDIA GeForce RTX 3090 + NVIDIA GeForce RTX 3090

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: no correctness check was run.

## Hardware and software

- GPU and VRAM: NVIDIA GeForce RTX 3090 + NVIDIA GeForce RTX 3090, 48 GiB VRAM
- CPU: AMD EPYC 9B45 128-Core Processor (128 cores, 256 threads); installed RAM: 178.2 GiB; storage: not measured
- OS: Linux 6.17.0-42-generic; GPU driver: 595.91.07; CUDA: 13.2
- Strata: project 0.1.45, commit d88a17d5; engine 0.1.45
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: qwen3.8-flash-next-iq3_xxs
- Context: 262144 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata --pack iq3_xxs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --mtp rt --max-context 262144 --kv int8 --kv-resident 32768 --vision --vram-reserve-mib 700 --pcie-frac 0.55 --spec-min-p 0.70 --pool-workers 64 --prefill-pipe 3072
```

## Method

Run by `python tools/strata_bench.py` 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), 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 | 125.8 (119.7-179.9) | 0.23 (0.22-1.00) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 2294.3 (2290.6-2299.3) | not measured | 1.77 (1.76-1.77) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 3272.7 (3270.7-3288.8) | not measured | 4.93 (4.91-4.93) |
| agent session, follow-up turns | 14587 (14573-14631) at the last turn | see below | up to 64 | 3 | see below | 139.4 (72.2-155.2) | 1.54 (0.96-1.71) |

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: 68.8 % (68.8-68.9); earlier-turn tokens read again per session: 3 (2-4) (earlier turns that the cache did not hold; the 13718 (13717-13719) new tokens of the files are read once); first turn (nothing cached) first token: 0.62 s (0.59-0.62).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 1052.4 | 0.593 | 114.8 |
| 1 | 3744 | 624 | 3120 | 1989.9 | 1.585 | 155.2 |
| 2 | 5277 | 3807 | 1470 | 1535.9 | 0.974 | 139.4 |
| 3 | 8710 | 5341 | 3369 | 1997.9 | 1.704 | 141.6 |
| 4 | 11636 | 8773 | 2863 | 1911.7 | 1.516 | 132.0 |
| 5 | 14587 | 11688 | 2899 | 1909.6 | 1.537 | 111.2 |

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: 63 s.