#6146 [Benchmark] AMD Radeon 890M (gfx1150, Windows) / Strata 0.2.13. qwen3.8-flash-next-unsloth-ud-iq2_xs
closed · @oleg-ka · 2 comentários · No GitHub
BenchmarksSetup & installAMD / HIPNVIDIA / CUDAModels & quantsSecurityWindows
Descrição
Measured 2026-11-16 one machine - 64GB RAM, engine 0.1.59 (0.2.13 strata). Built from sources in Windows. # Community benchmark on AMD Radeon(TM) 890M Graphics Measured on 2026-11-17 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(TM) 890M Graphics, 32 GiB VRAM - CPU: AMD Ryzen AI 9 HX PRO 370 w/ Radeon 890M (12 cores, 24 threads); installed RAM: 31.6 GiB; storage: CT2000E100SSD8 SSD/NVMe - OS: Windows 11 (10.0.26300); GPU driver: 32.0.32015.7029; CUDA (driver supports up to): not read - Strata: project 0.1.59, commit not a git checkout; engine 0.1.59 - Background workloads: not recorded (GPU power limit/draw read above) ## Model and configuration - Model and quantization: qwen3.8-flash-next-unsloth-ud-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 unsloth-ud-iq2_xs --native Qwen3.8-Flash-Next-UD-IQ2_XS-00001-of-00003.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 --resident-experts --resident-budget-gib 8 --vision --vram-reserve-mib 1000 --pool-workers 7 ``` Environment of the config: `STRATA_HIPBLASLT_TUNING=gfx1150-hipblaslt-100500.txt`, `STRATA_GDN_HEAD=1`, `STRATA_GDN_PP=2`, `STRATA_GDN_CONVL2=1`, `STRATA_GDN_NOY=1`, `STRATA_CVEC_FUSE=1`, `STRATA_PF_PAD=1`, `STRATA_Q8_PACKED=1`, `STRATA_Q6_PACKED=1`, `STRATA_MMVF_ROWS=1`, `STRATA_ATTN_LANECELL=1`, `STRATA_EXPERT_V2=1`, `STRATA_TSUM=1`, `STRATA_LFUSE=1`, `STRATA_GDN_SPLIT=1`, `STRATA_QFUSE=1`, `STRATA_PLE_BATCH=1`, `STRATA_SH_STREAM=1` ## 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 | 17.2 (14.5-23.4) | 1.43 (1.29-1.54) | | 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 178.1 (168.0-178.4) | not measured | 22.67 (22.65-24.04) | | 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 169.0 (166.0-171.9) | not measured | 95.14 (93.52-96.86) | | agent session, follow-up turns | 13565 (13530-13567) at the last turn | see below | up to 64 | 3 | see below | 17.3 (15.3-20.5) | 19.68 (12.14-23.31) | 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: 69.8 % (69.8-69.8); earlier-turn tokens read again per session: 5 (5-5) (earlier turns that the cache did not hold; the 12730 (12730-12730) new tokens of the files are read once); first turn (nothing cached) first token: 6.00 s (5.91-6.03). | Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s | | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | 0 | 609 | 0 | 609 | 103.0 | 6.032 | 16.2 | | 1 | 3832 | 627 | 3205 | 153.9 | 20.942 | 17.8 | | 2 | 5365 | 3895 | 1470 | 122.2 | 12.142 | 15.6 | | 3 | 8611 | 5428 | 3183 | 139.5 | 22.93 | 16.7 | | 4 | 10821 | 8653 | 2168 | 136.7 | 15.964 | 15.8 | | 5 | 13567 | 10858 | 2709 | 142.8 | 19.079 | 18.3 | 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 is not recorded. Total time of the suite: 786 s.
No site
Links install, modelos, releases.