#2250 Community benchmark on AMD Radeon 9600 XT (gfx1200) with 16 GB VRAM & 32 GB RAM

本文

# Community benchmark on AMD Radeon

Measured on 2026-10-13 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, 16 GiB VRAM
- CPU: AMD Ryzen 7 5700X 8-Core Processor (8 cores, 16 threads); installed RAM: 30.3 GiB; storage: not measured
- OS: Linux 7.0.0-38-generic; GPU driver: not read; CUDA: not read
- Strata: 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 (hip backend); thinking off for the suite; temperature 0

```text
strata --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 --spec-min-p 0.5 --mtp rt --max-context 65536 --kv int8 --resident-experts --control-vector-scaled Qwen3.8-Flash-Next-experimental-speed-projection.gguf:1.0 --control-vector-layer-range 4 44 --cvec-mode project --cvec-dir per-layer
```

Environment of the config: `STRATA_RESIDENT_PIN=0`

## 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 m
odel'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 | 37.6 (36.9-49.4) | 0.70 (0.64-0.79) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 905.8 (893.3-908.6) | not measured | 4.47 (4.46-4.53) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 967.4 (959.2-968.8) | not measured | 16.64 (16.61-16.78) |
| agent session, follow-up turns | 14445 (14422-14508) at the last turn | see below | up to 64 | 3 | see below | 41.7 (29.4-46.9) | 4.43 (2.74-5.04) |

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 (3-75) (earlier turns that the cache did not hold; the 13632 (13632-13635) new tokens of the files are read once); first turn (nothing cached) first token: 3.20 s (2.16-3.21).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 286.9 | 2.157 | 42.3 |
| 1 | 3728 | 648 | 3080 | 782.2 | 3.975 | 41.7 |
| 2 | 5261 | 3791 | 1470 | 544.5 | 2.736 | 39.4 |
| 3 | 8694 | 5314 | 3380 | 754.9 | 4.525 | 29.4 |
| 4 | 11577 | 8717 | 2860 | 718.2 | 4.022 | 36.5 |
| 5 | 14445 | 11599 | 2846 | 725.8 | 3.962 | 40.7 |

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: 195 s.