#2219 Community benchmark: 2x RTX 4070, E5-1630, 32GB, IQ1_M, 64K

Descrição

# Community benchmark on NVIDIA GeForce RTX 4070 + NVIDIA GeForce RTX 4070

Measured on 2026-10-12 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 4070 + NVIDIA GeForce RTX 4070, 24 GiB VRAM
- CPU: Intel(R) Xeon(R) CPU E5-1630 v3 @ 3.70GHz (4 cores, 8 threads); installed RAM: 31.2 GiB; storage: not measured
- OS: Linux 7.0.14-12-pve; GPU driver: 595.84; CUDA: 13.2
- <user>: project 0.1.43, commit 7695859a; 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: 65536 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
<user> --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 65536 --kv q4_0 --resident-experts --pcie-frac 0.55 --spec-min-p 0.70 --adapt-every 1 --adapt-swaps 80 --adapt-decay 0.97
```

## Method

Run by `python tools/strata_bench.py` against the HTTP API of a <user> 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 | 59.6 (55.3-77.2) | 0.48 (0.45-0.56) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 1652.4 (1496.0-1659.3) | not measured | 2.45 (2.44-2.71) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 2443.7 (2380.9-2450.6) | not measured | 6.60 (6.58-6.77) |
| agent session, follow-up turns | 14465 (14446-14474) at the last turn | see below | up to 64 | 3 | see below | 76.5 (59.9-84.9) | 2.03 (1.47-2.63) |

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 (3-4) (earlier turns that the cache did not hold; the 13632 (13632-13636) new tokens of the files are read once); first turn (nothing cached) first token: 1.39 s (1.39-1.45).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 428.3 | 1.445 | 38.4 |
| 1 | 3701 | 621 | 3080 | 1206.8 | 2.579 | 59.9 |
| 2 | 5234 | 3764 | 1470 | 882.4 | 1.692 | 60.8 |
| 3 | 8661 | 5281 | 3380 | 1299.5 | 2.63 | 60.7 |
| 4 | 11577 | 8717 | 2860 | 1221.3 | 2.372 | 81.3 |
| 5 | 14446 | 11600 | 2846 | 1299.7 | 2.221 | 83.2 |

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: 100 s.