#3468 [Benchmark] Community benchmark on AMD Radeon RX 6800 XTX + 48gb of RAM for 256k tokens of context

Description

# Community benchmark on AMD Radeon

Measured on 2026-10-20 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 3700X 8-Core Processor (8 cores, 16 threads); installed RAM: 46.9 GiB; storage: not measured
- OS: Linux 7.2.9-2-cachyos; GPU driver: 7.2.9-2-cachyos; CUDA: not read
- Strata: project 0.1.46.1, commit 9ae93f7b; engine 0.1.46.1
- Background workloads and power limits: not recorded

## Model and configuration

- Model and quantization: swift-1.5-iq3_xxs
- Context: 262144 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

## 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 | 32.0 (24.9-34.8) | 0.96 (0.86-1.05) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 344.5 (340.5-344.7) | not measured | 11.71 (11.70-11.84) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 430.7 (429.3-431.2) | not measured | 37.33 (37.28-37.45) |
| agent session, follow-up turns | 14698 (14683-14704) at the last turn | see below | up to 64 | 3 | see below | 37.1 (30.3-46.9) | 9.71 (7.26-10.78) |

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: 69.1 % (69.1-69.4); earlier-turn tokens read again per session: 143 (4-144) (earlier turns that the cache did not hold; the 13827 (13826-13828) new tokens of the files are read once); first turn (nothing cached) first token: 5.60 s (5.55-5.62).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 110.8 | 5.548 | 22.6 |
| 1 | 4085 | 620 | 3465 | 323.9 | 10.761 | 30.6 |
| 2 | 5618 | 4078 | 1540 | 207.9 | 7.453 | 31.9 |
| 3 | 9051 | 5611 | 3440 | 320.6 | 10.775 | 33.7 |
| 4 | 11821 | 9113 | 2708 | 285.0 | 9.547 | 43.4 |
| 5 | 14704 | 11885 | 2819 | 291.2 | 9.724 | 30.3 |

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: 390 s.