#2133 Community Benchmark: 5090 + Threadripper 9960X + Swift 1.5 IQ3_S

Description

# Community benchmark on NVIDIA GeForce RTX 5090

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 5090, 32 GiB VRAM
- CPU: AMD Ryzen Threadripper 9960X 24-Cores (24 cores, 48 threads); installed RAM: 63.4 GiB; storage: not measured
- OS: Windows 11 (10.0.26200); GPU driver: 617.42; 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: swift-1.5-iq3_s
- Context: 131072 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata.exe --pack swift-iq3_s --native Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --ple-gguf Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --mtp rt --max-context 131072 --kv int8 --vision --vram-reserve-mib 700 --pcie-frac 0.35 --spec-min-p 0.50
```

## 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 | 185.6 (142.5-226.2) | 0.20 (0.15-0.23) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 5095.7 (4950.1-5120.4) | not measured | 0.80 (0.80-0.83) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 6639.4 (6561.0-6642.7) | not measured | 2.45 (2.45-2.48) |
| agent session, follow-up turns | 14427 (14426-14468) at the last turn | see below | up to 64 | 3 | see below | 184.7 (142.2-257.5) | 0.75 (0.65-0.88) |

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 % (68.4-68.8); earlier-turn tokens read again per session: 74 (3-144) (earlier turns that the cache did not hold; the 13635 (13635-13635) new tokens of the files are read once); first turn (nothing cached) first token: 0.45 s (0.43-0.48).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 1462.2 | 0.448 | 140.5 |
| 1 | 3715 | 633 | 3082 | 3858.3 | 0.825 | 190.7 |
| 2 | 5248 | 3778 | 1470 | 2221.6 | 0.69 | 184.7 |
| 3 | 8692 | 5241 | 3451 | 4050.9 | 0.881 | 170.3 |
| 4 | 11595 | 8735 | 2860 | 3675.6 | 0.816 | 214.2 |
| 5 | 14468 | 11622 | 2846 | 3791.1 | 0.779 | 180.3 |

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: 36 s.