#2128 Community Benchmark: RX 7900xtx w/ AMD EPYC 7K62 48-Core

Description

# Community benchmark on AMD Radeon RX 7900xtx

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: AMD Radeon RX 7900XTX, 24 GiB VRAM
- CPU: AMD EPYC 7K62 48-Core Processor (48 cores, 96 threads); installed RAM: 184.3 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: swift-1.5-iq3_s
- Context: 524288 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 --long` 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 | 63.0 (51.7-72.2) | 0.71 (0.62-0.78) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 1281.9 (1266.2-1281.9) | not measured | 3.16 (3.16-3.20) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 1969.7 (1963.7-1972.1) | not measured | 8.19 (8.18-8.22) |
| 32K prompt, fresh | 32049 (32049-32049) | 0 (0-0) | 24 cap | 3 | 2047.7 (2015.7-2048.4) | not measured | 15.70 (15.70-15.95) |
| agent session, follow-up turns | 14480 (14458-14487) at the last turn | see below | up to 64 | 3 | see below | 69.6 (56.9-106.4) | 2.70 (2.10-3.30) |

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 % (68.5-68.8); earlier-turn tokens read again per session: 73 (3-144) (earlier turns that the cache did not hold; the 13635 (13635-13636) new tokens of the files are read once); first turn (nothing cached) first token: 2.28 s (1.95-2.28).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 316.9 | 1.949 | 60.1 |
| 1 | 3715 | 633 | 3082 | 1097.7 | 2.842 | 56.9 |
| 2 | 5248 | 3708 | 1540 | 653.3 | 2.389 | 61.1 |
| 3 | 8692 | 5241 | 3451 | 1057.5 | 3.299 | 67.2 |
| 4 | 11609 | 8749 | 2860 | 1031.7 | 2.81 | 81.8 |
| 5 | 14480 | 11634 | 2846 | 1041.7 | 2.771 | 67.6 |

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: 171 s.