#2263 Community benchmark on NVIDIA GeForce RTX 4090

Beschreibung

# Community benchmark on NVIDIA GeForce RTX 4090

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 GeForce RTX 4090, 24 GiB VRAM
- CPU: Intel(R) Core(TM) i9-14900KF (24 cores, 32 threads); installed RAM: 127.8 GiB; storage: not measured
- OS: Windows 11 (10.0.26200); GPU driver: 616.92; CUDA: not read
- Strata: project 0.1.43, commit not a git checkout; engine 0.1.43
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: qwen3.8-flash-next-coder-iq1_m
- Context: 393216 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata.exe --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-cache auto --prefill auto --spec 4 --mtp rt --max-context 393216 --rope-scaling yarn --rope-scale 1.5 --kv int8 --kv-resident 32768 --vision --vram-reserve-mib 700 --pcie-frac 0.20 --spec-min-p 0.70
```

## 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 | 89.1 (84.6-97.3) | 0.38 (0.31-0.41) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 3516.5 (3444.5-3541.3) | not measured | 1.17 (1.15-1.20) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 4572.7 (4498.2-4596.0) | not measured | 3.54 (3.52-3.60) |
| agent session, follow-up turns | 14463 (14439-14486) at the last turn | see below | up to 64 | 3 | see below | 105.6 (86.0-124.9) | 1.14 (0.99-1.25) |

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 % (68.7-68.8); earlier-turn tokens read again per session: 4 (4-4) (earlier turns that the cache did not hold; the 13634 (13633-13634) new tokens of the files are read once); first turn (nothing cached) first token: 0.80 s (0.79-0.83).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 793.1 | 0.786 | 93.8 |
| 1 | 3714 | 631 | 3083 | 2542.7 | 1.245 | 86.0 |
| 2 | 5247 | 3778 | 1469 | 1517.1 | 1.0 | 109.6 |
| 3 | 8662 | 5282 | 3380 | 2796.6 | 1.246 | 97.8 |
| 4 | 11582 | 8722 | 2860 | 2510.1 | 1.166 | 104.6 |
| 5 | 14486 | 11640 | 2846 | 2520.8 | 1.157 | 107.4 |

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: 61 s.