#2845 Community benchmark: 2x Tesla V100 16 GB (SXM2 + PCIe), i9-9820X, Qwen3.8-Flash-Next IQ3_S

Descrição

First dual-V100-16GB entry I think - layer split across V100-SXM2-16GB + V100-PCIE-16GB with KV streaming, `STRATA_SM70_TABLE=1`. Ran `./bench.sh` (suite 1) against the running server, 2026-10-17.

Context: this box also runs a production agent workload; daily logs show decode weighted ~74-77 tok/s with 94% prompt-cache reuse, consistent with the suite agent-session figures.

# Community benchmark on Tesla V100-SXM2-16GB + Tesla V100-PCIE-16GB

Measured on 2026-10-17 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: Tesla V100-SXM2-16GB + Tesla V100-PCIE-16GB, 32 GiB VRAM
- CPU: Intel(R) Core(TM) i9-9820X CPU @ 3.30GHz (10 cores, 20 threads); installed RAM: 123.2 GiB; storage: not measured
- OS: Linux 7.0.0-38-generic; GPU driver: 580.178.04; CUDA: 13.0
- Strata: project 0.1.44.1, commit 296b59d2; 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: 262144 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 262144 --kv int8 --kv-resident 32768 --remote-expert-opt --vram-reserve-mib 768 --vram-reserve-later-mib 1105 --spec-min-p 0.5 --vision
```

Environment of the config: `STRATA_SM70_TABLE=1`

## 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 | 66.5 (54.1-79.4) | 0.49 (0.42-0.57) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 821.3 (805.1-821.5) | not measured | 4.92 (4.92-5.02) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 1579.4 (1569.7-1579.5) | not measured | 10.20 (10.20-10.27) |
| agent session, follow-up turns | 14567 (14555-14581) at the last turn | see below | up to 64 | 3 | see below | 71.0 (62.1-88.7) | 5.14 (4.68-5.46) |

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: 68.6 % (68.6-68.8); earlier-turn tokens read again per session: 74 (4-74) (earlier turns that the cache did not hold; the 13728 (13728-13728) new tokens of the files are read once); first turn (nothing cached) first token: 3.33 s (3.31-3.35).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 185.5 | 3.306 | 53.0 |
| 1 | 3743 | 623 | 3120 | 592.1 | 5.296 | 62.1 |
| 2 | 5276 | 3736 | 1540 | 325.2 | 4.763 | 64.5 |
| 3 | 8720 | 5339 | 3381 | 622.7 | 5.46 | 70.3 |
| 4 | 11628 | 8766 | 2862 | 549.1 | 5.245 | 76.8 |
| 5 | 14555 | 11656 | 2899 | 560.9 | 5.203 | 68.4 |

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: 175 s.