#2876 benchmark rx9070xt

Beschreibung

# Community benchmark on AMD Radeon RX 9070 XT

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: AMD Radeon RX 9070 XT, 16 GiB VRAM
- CPU: AMD Ryzen 9 5900XT 16-Core Processor (16 cores, 32 threads); installed RAM: 63.9 GiB; storage: not measured
- OS: Windows 11 (10.0.26200); GPU driver: 32.0.23027.2005; CUDA: not read
- Strata: project 0.1.45, commit 477ac13c; engine 0.1.45
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: qwen3.8-flash-next-iq2_xs
- Context: 131072 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged (hip backend); thinking off for the suite; temperature 0

```text
strata.exe --pack iq2_xs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 131072 --kv int8 --kv-resident 32768
```

Environment of the config: `STRATA_HIPBLASLT_TUNING=gfx1201-hipblaslt-100500.txt`

## 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 | 80.3 (68.4-87.7) | 0.38 (0.34-0.43) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 810.4 (809.0-820.3) | not measured | 4.99 (4.92-5.00) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 958.1 (955.5-962.0) | not measured | 16.81 (16.73-16.84) |
| agent session, follow-up turns | 14651 (14651-14651) at the last turn | see below | up to 64 | 2 | see below | 121.6 (67.0-130.6) | 4.72 (4.07-4.89) |

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: 69.0 % (69.0-69.0); earlier-turn tokens read again per session: 4 (0-5) (earlier turns that the cache did not hold; the 13722 (0-13722) new tokens of the files are read once); first turn (nothing cached) first token: 2.24 s (2.20-2.27).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 270.5 | 2.271 | 74.3 |
| 1 | 3796 | 672 | 3124 | 665.7 | 4.718 | 100.6 |
| 2 | 5329 | 3859 | 1470 | 363.0 | 4.074 | 116.4 |
| 3 | 8762 | 5393 | 3369 | 693.1 | 4.886 | 119.2 |
| 4 | 11688 | 8825 | 2863 | 612.1 | 4.722 | 120.9 |
| 5 | 14651 | 11751 | 2900 | 616.2 | 4.733 | 67.0 |

### Skipped and failed

- failed agent: no token came back

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: 152 s.

[results.json](https://github.com/user-attachments/files/33593565/results.json)