Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

GPU 性能基准测试 · yangge521 | DecimalAI

ai mlby yangge521

GPU 性能基准测试

yangge521/gpu-benchmark

Source yangge521/dev-skills-zh ↗License MIT

GPU 性能基准测试工具包，涵盖训练速度对比、推理延迟分析、极限 Batch Size 测试

SkillScore91Maintained

Automatic

AI Engineer pack

Last verified 8/3/2026

What it changes

Passed 20 of 22 test cases with the skill, and 16 of 22 without it.

Without skill73%

With skill91%

Checks it passed that the model alone missed

States NVIDIA T4 NMS latency is 0.9ms
States MTT S4000 NMS latency is 13.7ms
Cites NVIDIA T4 training time as 2.053 hours

and 3 more in the full report

4 fixed16 passed either way2 still failing0 broke

Pass-rate lift: +18 pp
Tokens: +21%
Agent turns: 0%
Response time: -31%

Model gemini-3.6-flashMeasured 8/3/2026v120 of 22 cases passed with skill

Tokens, turns and time describe resource use, not a dollar price.

Use this skill

Save it into your project's skills folder. Claude Code picks it up in its next session.

curl -fsSL --create-dirs -o .claude/skills/yangge521-gpu-benchmark/SKILL.md https://app.decimal.ai/s/yangge521-gpu-benchmark/SKILL.md

Want it pinned and updatable? pip install "decimalai>=0.15", then decimalai skills add yangge521-gpu-benchmark, and start Claude Code with claude --plugin-dir "$(git rev-parse --show-toplevel)/.agents"

1 · One-time — install the CLI (docs ↗)

pip install "decimalai>=0.15"

2 · Get the skill — writes .agents/skills/yangge521-gpu-benchmark/ and pins it in .decimal/skills.lock

decimalai skills add yangge521-gpu-benchmark

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/yangge521-gpu-benchmark/SKILL.md https://app.decimal.ai/s/yangge521-gpu-benchmark/SKILL.md

Raw file: /s/yangge521-gpu-benchmark/SKILL.md ↗ · machine-readable summary at /s/yangge521-gpu-benchmark.json ↗

Safety checks: 3 of 3 passed

Code scan CleanIntent PassedContent Clean

What's inside

The instructions your model reads when the skill loads.

SKILL.md106 linessingle file

GPU 性能基准测试

概述

提供 GPU 性能基准测试的完整方案，包括训练速度对比、推理延迟分解分析、极限 Batch Size 探测。基于 MTT S4000 (48GB) 与 NVIDIA T4 (16GB) 的真实对比测试数据。

使用场景

GPU 选型评估：对比不同 GPU 的性价比
训练性能对比：评估不同 GPU 的训练吞吐
推理延迟瓶颈定位：分解各阶段耗时
最大 Batch Size 探测：确定最优批处理大小
跨 GPU 平台基准：建立性能基准线
驱动/固件升级验证：升级前后性能对比

真实测试数据

1. 训练速度对比（YOLO11n, 100 epochs, COCO128）

GPU	训练时间	平均 epoch	显存占用
NVIDIA T4 (16GB)	2.053h	73.9s	3.2GB
MTT S4000 (48GB)	1.919h	69.1s	3.5GB

S4000 训练速度比 T4 快约 7%

2. 推理延迟分解（YOLO11n, batch=1）

阶段	NVIDIA T4	MTT S4000	差异
前向推理	1.3ms	1.2ms	S4000 略快
后处理 NMS	0.9ms	13.7ms	S4000 慢 15x
总延迟	2.2ms	14.9ms	T4 更快

S4000 后处理（NMS）耗时 13.7ms，是 T4 的 15 倍，为主要瓶颈

3. 极限 Batch Size 测试

GPU	显存	最大 Batch Size	OOM 临界点

Track record

Benchmark

20 / 22

test cases passed, vs 16 / 22 without

Adoption

Users: 0
Times used: 0

Safety and source details

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/1/2026

View 15 scanner checks

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

yangge521 / dev-skills-zh

skills/ai-engineering/gpu-benchmark/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v1

More skills from the same source and nearby in the registry, with their headline benchmark lift.

Ready to supercharge your AI agents?

What's inside

GPU 性能基准测试

概述

使用场景

真实测试数据

1. 训练速度对比（YOLO11n, 100 epochs, COCO128）

2. 推理延迟分解（YOLO11n, batch=1）

3. 极限 Batch Size 测试

Track record

Benchmark

Adoption

SkillSafety

Source

Related skills