MiniCPM-V-4_7-35B-A3B-Q4_K_M Benchmark

[Haroldyolo]

Community benchmark on NVIDIA GeForce RTX 4070 Ti SUPER

Measured on 2026-10-21 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: NVIDIA GeForce RTX 4070 Ti SUPER, 16 GiB VRAM
• CPU: AMD Ryzen 7 7800X3D 8-Core Processor (8 cores, 16 threads); installed RAM: 30.9 GiB; storage: not measured
• OS: Windows 11 (10.0.26200); GPU driver: 610.88; CUDA: not read
• PolyStrata: project 0.1.45, commit a418c11; engine 0.1.45
• Background workloads and power limits: not recorded

Model and configuration

• Model and quantization: minicpm-v-4.7-35b-a3b-q4_k_m
• Context: 65536 tokens; parallel slots: 1
• Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

strata-poly.exe --model MiniCPM-V-4_7-35B-A3B-Q4_K_M.gguf --max-context 65536 --expert-profile expert-profile.txt

Method

Run by python tools/strata_bench.py against the HTTP API of a PolyStrata 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	134.2 (94.5-142.4)	0.24 (0.21-0.29)
4K prompt, fresh	4019 (4019-4019)	0 (0-0)	24 cap	3	1997.6 (1982.2-2010.0)	not measured	2.03 (2.00-2.05)
16K prompt, fresh	16058 (16058-16058)	0 (0-0)	24 cap	3	3179.6 (3077.0-3184.4)	not measured	5.07 (5.06-5.24)
agent session, follow-up turns	15156 (15155-15157) at the last turn	see below	up to 64	3	see below	123.3 (102.9-150.4)	2.21 (1.63-2.52)

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: 69.2 % (69.2-69.2); earlier-turn tokens read again per session: 5 (5-5) (earlier turns that the cache did not hold; the 14334 (14334-14334) new tokens of the files are read once); first turn (nothing cached) first token: 0.82 s (0.80-0.85).

Turn (run 1)	Prompt tokens	Reused	Read	Prompt tok/s	First token (s)	Decode tok/s
0	609	0	609	717.7	0.852	123.5
1	4178	619	3559	1498.1	2.398	117.5
2	5824	4227	1597	986.3	1.646	116.8
3	9250	5878	3372	1365.3	2.499	118.1
4	12220	9298	2922	1358.2	2.16	102.9
5	15157	12268	2889	1331.8	2.178	103.8

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: 82 s.

[VecSzn]

Thanks same PC as the Tiel and Qwen3.6 runs, so it should compare pretty well. Added it to docs/COMMUNITY_BENCHMARKS.md right next to those rows.