| <!DOCTYPE html> |
| <html lang="ko"> |
| <head> |
| <meta charset="UTF-8"> |
| <meta name="viewport" content="width=device-width, initial-scale=1.0"> |
| <title>GPU/LLM 전력 실험 결과 대시보드</title> |
| <script src="https://cdn.plot.ly/plotly-2.35.2.min.js" charset="utf-8"></script> |
| <style> |
| * { margin: 0; padding: 0; box-sizing: border-box; } |
| body { |
| font-family: 'Segoe UI', 'Apple SD Gothic Neo', 'Malgun Gothic', sans-serif; |
| background: #0f172a; color: #e2e8f0; |
| line-height: 1.6; padding: 1rem 1.5rem; |
| } |
| .container { max-width: 1800px; margin: 0 auto; } |
| |
| h1 { |
| font-size: 2rem; font-weight: 800; margin-bottom: 0.25rem; |
| background: linear-gradient(135deg, #38bdf8, #818cf8); |
| -webkit-background-clip: text; -webkit-text-fill-color: transparent; |
| } |
| .subtitle { color: #94a3b8; font-size: 0.95rem; margin-bottom: 1.5rem; } |
| |
| .meta-bar { |
| display: flex; gap: 1.5rem; background: #1e293b; border-radius: 12px; |
| padding: 0.75rem 1.25rem; margin-bottom: 1.5rem; flex-wrap: wrap; |
| } |
| .meta-item { font-size: 0.8rem; color: #94a3b8; } |
| .meta-item strong { color: #e2e8f0; } |
| |
| .tabs { display: flex; gap: 0.5rem; margin-bottom: 1.5rem; flex-wrap: wrap; } |
| .tab-btn { |
| padding: 8px 20px; border-radius: 8px; border: 1px solid #334155; |
| background: #1e293b; color: #94a3b8; cursor: pointer; |
| font-size: 0.85rem; font-weight: 600; transition: all 0.2s; |
| } |
| .tab-btn:hover { border-color: #38bdf8; color: #e2e8f0; } |
| .tab-btn.active { background: #38bdf8; color: #0f172a; border-color: #38bdf8; } |
| .tab-content { display: none; } |
| .tab-content.active { display: block; } |
| |
| .exp-card { |
| background: #1e293b; border-radius: 12px; border: 1px solid #334155; |
| padding: 1.5rem; margin-bottom: 1.5rem; |
| transition: border-color 0.2s; |
| } |
| .exp-card:hover { border-color: #38bdf8; } |
| .exp-header { display: flex; align-items: baseline; gap: 0.75rem; margin-bottom: 0.5rem; } |
| .exp-number { |
| font-size: 0.7rem; font-weight: 700; color: #38bdf8; |
| background: rgba(56,189,248,0.1); padding: 2px 8px; border-radius: 6px; |
| white-space: nowrap; |
| } |
| .exp-title { font-size: 1.15rem; font-weight: 600; color: #f1f5f9; } |
| .exp-desc { color: #94a3b8; font-size: 0.85rem; margin-bottom: 0.75rem; } |
| .exp-variable { |
| font-size: 0.75rem; color: #a78bfa; margin-bottom: 1rem; |
| display: inline-block; background: rgba(167,139,250,0.1); |
| padding: 2px 10px; border-radius: 6px; |
| } |
| |
| |
| .comp-table { |
| width: 100%; border-collapse: collapse; font-size: 0.8rem; |
| margin-bottom: 1rem; overflow-x: auto; display: block; |
| } |
| .comp-table th { |
| text-align: left; padding: 0.5rem 0.6rem; color: #94a3b8; |
| border-bottom: 2px solid #334155; font-weight: 600; font-size: 0.7rem; |
| text-transform: uppercase; letter-spacing: 0.3px; |
| background: #162032; position: sticky; top: 0; white-space: nowrap; |
| } |
| .comp-table td { |
| padding: 0.4rem 0.6rem; border-bottom: 1px solid #0f172a; white-space: nowrap; |
| } |
| .comp-table tr:hover td { background: rgba(56,189,248,0.05); } |
| .comp-table .num { text-align: right; font-variant-numeric: tabular-nums; } |
| .comp-table .delta { color: #64748b; font-size: 0.7rem; margin-left: 4px; } |
| .comp-table .baseline-row td { background: rgba(56,189,248,0.08); font-weight: 600; } |
| |
| |
| .comparison-chart-wrapper { |
| margin: 1rem 0; background: #162032; border-radius: 8px; |
| border: 1px solid #334155; padding: 0.5rem; overflow: hidden; |
| } |
| .comparison-chart { width: 100%; height: 450px; } |
| |
| |
| .graph-grid { |
| display: grid; |
| grid-template-columns: 1fr; |
| gap: 1rem; margin-top: 1rem; |
| } |
| .graph-item { |
| background: #162032; border-radius: 8px; border: 1px solid #334155; |
| overflow: hidden; |
| } |
| .graph-label { |
| padding: 6px 12px; font-size: 0.75rem; color: #94a3b8; |
| border-bottom: 1px solid #334155; font-weight: 600; |
| white-space: nowrap; overflow: hidden; text-overflow: ellipsis; |
| } |
| .graph-item iframe { |
| width: 100%; height: 520px; border: none; background: #0f172a; |
| } |
| |
| .section-divider { |
| display: flex; align-items: center; gap: 1rem; margin: 2rem 0 1rem; |
| } |
| .section-divider h2 { |
| font-size: 1.3rem; font-weight: 700; color: #f1f5f9; white-space: nowrap; |
| } |
| .section-divider::after { |
| content: ''; flex: 1; height: 1px; background: #334155; |
| } |
| |
| .collapsible-toggle { |
| cursor: pointer; user-select: none; color: #38bdf8; |
| font-size: 0.8rem; font-weight: 600; margin-top: 0.5rem; |
| display: inline-block; |
| } |
| .collapsible-toggle:hover { text-decoration: underline; } |
| .collapsible-content { margin-top: 0.75rem; } |
| |
| .empty-msg { color: #64748b; font-style: italic; padding: 1rem 0; } |
| |
| .footer { |
| margin-top: 2rem; padding-top: 1rem; border-top: 1px solid #334155; |
| color: #64748b; font-size: 0.75rem; text-align: center; |
| } |
| |
| .glossary { |
| background: #1e293b; border-radius: 12px; border: 1px solid #334155; |
| padding: 1.25rem; margin-bottom: 1.5rem; |
| } |
| .glossary h3 { font-size: 1rem; margin-bottom: 0.5rem; color: #f1f5f9; } |
| .glossary dl { display: grid; grid-template-columns: 160px 1fr; gap: 0.2rem 1rem; } |
| .glossary dt { font-weight: 600; color: #38bdf8; font-size: 0.8rem; } |
| .glossary dd { color: #94a3b8; font-size: 0.8rem; } |
| </style> |
| </head> |
| <body> |
| <div class="container"> |
| <h1>GPU / LLM 전력 실험 결과 대시보드</h1> |
| <p class="subtitle">AI 데이터센터 전력 프로파일 분석 — SEND Lab 전송 패키지</p> |
|
|
| <div class="meta-bar"> |
| <div class="meta-item">GPU(비전) 실험: <strong>11그룹 / 30개 run</strong></div> |
| <div class="meta-item">LLM 실험: <strong>12그룹 / 37개 run</strong></div> |
| <div class="meta-item">하드웨어: <strong>NVIDIA RTX A6000 / PRO 6000</strong></div> |
| <div class="meta-item">샘플링: <strong>10 ms (NVML)</strong></div> |
| <div class="meta-item">생성일: <strong>2026-02-19 16:15</strong></div> |
| </div> |
|
|
| <div class="glossary"> |
| <h3>📖 용어 설명</h3> |
| <dl> |
| <dt>Idle 평균 (W)</dt> |
| <dd>실험 시작 전 대기 상태(GPU 로딩 완료, 연산 없음)의 평균 전력입니다.</dd> |
| <dt>학습 평균/피크 (W)</dt> |
| <dd>모델 학습(순전파+역전파) 중 평균 및 최대 전력입니다.</dd> |
| <dt>추론 평균/피크 (W)</dt> |
| <dd>추론 요청 처리 중(LLM: Decode 단계) 평균 및 최대 전력입니다.</dd> |
| <dt>Decode Token/s (↑)</dt> |
| <dd>LLM이 1초당 생성한 토큰 수입니다. 높을수록 좋습니다.</dd> |
| <dt>J/Token (↓)</dt> |
| <dd>출력 토큰 1개당 소비한 에너지(줄)입니다. 낮을수록 효율적입니다.</dd> |
| <dt>(+X.XX) / (-X.XX)</dt> |
| <dd>기준 조건(baseline) 대비 변화량입니다. 기준 행은 <span style="color:#38bdf8">하이라이트</span> 표시됩니다.</dd> |
| </dl> |
| </div> |
|
|
| <div class="tabs"> |
| <button class="tab-btn active" onclick="switchTab('gpu')">🖥️ GPU (비전) 실험</button> |
| <button class="tab-btn" onclick="switchTab('llm')">🤖 LLM 실험</button> |
| </div> |
|
|
| <div id="tab-gpu" class="tab-content active"> |
| <div class="section-divider"><h2>🖥️ GPU (비전 모델) 실험 결과</h2></div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#00</span> |
| <span class="exp-title">기준 실험 (Baseline)</span> |
| </div> |
| <p class="exp-desc">전력 제어 없이 표준 조건으로 실행한 기준 실험입니다.</p> |
| <span class="exp-variable">실험 변수: 없음 (대조군)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.7 <span class="delta"></span></td><td class="num">134.3 <span class="delta"></span></td><td class="num">178.0 <span class="delta"></span></td><td class="num">78.1 <span class="delta"></span></td><td class="num">79.4 <span class="delta"></span></td><td class="num">29670 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_0')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_0" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/00_baseline_reference/fixed_resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#01</span> |
| <span class="exp-title">배치 크기 변화 실험</span> |
| </div> |
| <p class="exp-desc">배치 크기(16/64/128)에 따른 GPU 전력 소비 변화를 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 배치 크기 (16 / 64 / 128)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet50_cifar10_gpu0_bs128_ep1_req10000_nocap_fixed</td><td class="num">51.8 <span class="delta">(-0.659)</span></td><td class="num">138.1 <span class="delta">(+4.684)</span></td><td class="num">211.8 <span class="delta">(+38.582)</span></td><td class="num">78.2 <span class="delta">(-0.166)</span></td><td class="num">79.5 <span class="delta">(-0.095)</span></td><td class="num">29205 <span class="delta">(-526.306)</span></td></tr><tr><td>resnet50_cifar10_gpu0_bs16_ep1_req10000_nocap_fixed</td><td class="num">53.3 <span class="delta">(+0.836)</span></td><td class="num">139.9 <span class="delta">(+6.482)</span></td><td class="num">152.2 <span class="delta">(-20.997)</span></td><td class="num">77.9 <span class="delta">(-0.461)</span></td><td class="num">79.3 <span class="delta">(-0.337)</span></td><td class="num">32440 <span class="delta">(+2708.031)</span></td></tr><tr class="baseline-row"><td>resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.5 <span class="delta"></span></td><td class="num">133.4 <span class="delta"></span></td><td class="num">173.2 <span class="delta"></span></td><td class="num">78.3 <span class="delta"></span></td><td class="num">79.6 <span class="delta"></span></td><td class="num">29732 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_1" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["resnet50_cifar10_bs128_nocap_fixed", "resnet50_cifar10_bs16_nocap_fixed", "resnet50_cifar10_bs64_nocap_fixed"]; |
| var barTrace = { |
| x: labels, y: [138.1, 139.9, 133.4], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [78.2, 77.9, 78.3], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_1', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_1')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_1" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs128_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/01_batch_sweep_resnet50/fixed_resnet50_cifar10_gpu0_bs128_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs16_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/01_batch_sweep_resnet50/fixed_resnet50_cifar10_gpu0_bs16_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/01_batch_sweep_resnet50/fixed_resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#02</span> |
| <span class="exp-title">운영자 제어 시뮬레이션</span> |
| </div> |
| <p class="exp-desc">비균일 추론 요청 패턴에서의 전력 프로파일을 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 가변 추론 간격</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep5_req60_nocap_variable</td><td class="num">53.7 <span class="delta"></span></td><td class="num">128.5 <span class="delta"></span></td><td class="num">152.2 <span class="delta"></span></td><td class="num">78.2 <span class="delta"></span></td><td class="num">79.3 <span class="delta"></span></td><td class="num">34124 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_2')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_2" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep5_req60_nocap_variable</div> |
| <iframe src="gpu_power_experiment/02_operator_control/fixed_resnet18_cifar10_gpu0_bs64_ep5_req60_nocap_variable/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#03</span> |
| <span class="exp-title">전력 상한 (Power Cap) 실험</span> |
| </div> |
| <p class="exp-desc">nvidia-smi 전력 제한(345W/460W/575W)이 성능과 전력에 미치는 영향을 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 전력 상한 (345W / 460W / 575W)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_cap345W_fixed</td><td class="num">9.9 <span class="delta">(-42.174)</span></td><td class="num">111.6 <span class="delta">(+9.599)</span></td><td class="num">148.0 <span class="delta">(-1.972)</span></td><td class="num">77.9 <span class="delta">(+0.084)</span></td><td class="num">78.8 <span class="delta">(-0.122)</span></td><td class="num">27225 <span class="delta">(-1655.702)</span></td></tr><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_cap460W_fixed</td><td class="num">9.4 <span class="delta">(-42.712)</span></td><td class="num">118.1 <span class="delta">(+16.059)</span></td><td class="num">147.3 <span class="delta">(-2.643)</span></td><td class="num">78.1 <span class="delta">(+0.189)</span></td><td class="num">79.3 <span class="delta">(+0.345)</span></td><td class="num">27206 <span class="delta">(-1675.326)</span></td></tr><tr class="baseline-row"><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_cap575W_fixed</td><td class="num">52.1 <span class="delta"></span></td><td class="num">102.0 <span class="delta"></span></td><td class="num">150.0 <span class="delta"></span></td><td class="num">77.9 <span class="delta"></span></td><td class="num">79.0 <span class="delta"></span></td><td class="num">28881 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_3" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["resnet18_cifar10_bs64_cap345W_fixed", "resnet18_cifar10_bs64_cap460W_fixed", "resnet18_cifar10_bs64_cap575W_fixed"]; |
| var barTrace = { |
| x: labels, y: [111.6, 118.1, 102.0], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [77.9, 78.1, 77.9], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_3', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_3')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_3" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_cap345W_fixed</div> |
| <iframe src="gpu_power_experiment/03_control_cap_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_cap345W_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_cap460W_fixed</div> |
| <iframe src="gpu_power_experiment/03_control_cap_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_cap460W_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_cap575W_fixed</div> |
| <iframe src="gpu_power_experiment/03_control_cap_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_cap575W_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#04</span> |
| <span class="exp-title">단계적 전력 램프 실험</span> |
| </div> |
| <p class="exp-desc">전력 제한을 단계적으로 변화시켜 GPU 적응 동작을 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 램프 활성화</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_ramp_fixed</td><td class="num">54.1 <span class="delta"></span></td><td class="num">105.8 <span class="delta"></span></td><td class="num">148.3 <span class="delta"></span></td><td class="num">77.9 <span class="delta"></span></td><td class="num">78.9 <span class="delta"></span></td><td class="num">29572 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_4')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_4" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_ramp_fixed</div> |
| <iframe src="gpu_power_experiment/04_control_ramp/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_ramp_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#05</span> |
| <span class="exp-title">추론 패턴 비교 실험</span> |
| </div> |
| <p class="exp-desc">고정/가변/버스트 세 가지 추론 스케줄링 패턴의 전력 영향을 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 추론 패턴 (fixed / variable / burst)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr class="baseline-row"><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_burst</td><td class="num">51.9 <span class="delta"></span></td><td class="num">102.0 <span class="delta"></span></td><td class="num">151.7 <span class="delta"></span></td><td class="num">78.4 <span class="delta"></span></td><td class="num">79.7 <span class="delta"></span></td><td class="num">28671 <span class="delta"></span></td></tr><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.4 <span class="delta">(+1.513)</span></td><td class="num">100.1 <span class="delta">(-1.915)</span></td><td class="num">149.9 <span class="delta">(-1.777)</span></td><td class="num">78.3 <span class="delta">(-0.171)</span></td><td class="num">79.4 <span class="delta">(-0.301)</span></td><td class="num">28981 <span class="delta">(+310.083)</span></td></tr><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_variable</td><td class="num">52.3 <span class="delta">(+0.360)</span></td><td class="num">103.6 <span class="delta">(+1.544)</span></td><td class="num">150.7 <span class="delta">(-1.053)</span></td><td class="num">78.0 <span class="delta">(-0.441)</span></td><td class="num">79.2 <span class="delta">(-0.444)</span></td><td class="num">28852 <span class="delta">(+180.841)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_5" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["resnet18_cifar10_bs64_nocap_burst", "resnet18_cifar10_bs64_nocap_fixed", "resnet18_cifar10_bs64_nocap_variable"]; |
| var barTrace = { |
| x: labels, y: [102.0, 100.1, 103.6], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [78.4, 78.3, 78.0], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_5', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_5')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_5" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_burst</div> |
| <iframe src="gpu_power_experiment/05_pattern_fixed_var_burst/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_burst/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/05_pattern_fixed_var_burst/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_variable</div> |
| <iframe src="gpu_power_experiment/05_pattern_fixed_var_burst/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_variable/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#06</span> |
| <span class="exp-title">모델 아키텍처별 전력 비교</span> |
| </div> |
| <p class="exp-desc">10개 비전 모델 아키텍처가 생성하는 서로 다른 전력 시그니처를 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 모델 아키텍처 (10종)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>convnext_tiny_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.6 <span class="delta">(-0.100)</span></td><td class="num">325.6 <span class="delta">(+190.942)</span></td><td class="num">514.4 <span class="delta">(+340.990)</span></td><td class="num">79.1 <span class="delta">(+1.131)</span></td><td class="num">83.4 <span class="delta">(+3.834)</span></td><td class="num">42655 <span class="delta">(+12997.676)</span></td></tr><tr><td>densenet121_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.7 <span class="delta">(-0.091)</span></td><td class="num">378.6 <span class="delta">(+243.882)</span></td><td class="num">436.7 <span class="delta">(+263.225)</span></td><td class="num">79.2 <span class="delta">(+1.171)</span></td><td class="num">83.3 <span class="delta">(+3.767)</span></td><td class="num">43719 <span class="delta">(+14061.848)</span></td></tr><tr><td>efficientnet_b0_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.2 <span class="delta">(-0.546)</span></td><td class="num">310.7 <span class="delta">(+176.043)</span></td><td class="num">401.6 <span class="delta">(+228.123)</span></td><td class="num">78.9 <span class="delta">(+0.901)</span></td><td class="num">81.4 <span class="delta">(+1.868)</span></td><td class="num">38061 <span class="delta">(+8403.014)</span></td></tr><tr><td>mobilenetv2_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">54.0 <span class="delta">(+0.246)</span></td><td class="num">103.0 <span class="delta">(-31.713)</span></td><td class="num">138.9 <span class="delta">(-34.541)</span></td><td class="num">78.0 <span class="delta">(-0.018)</span></td><td class="num">79.4 <span class="delta">(-0.195)</span></td><td class="num">29468 <span class="delta">(-189.919)</span></td></tr><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.9 <span class="delta">(-0.883)</span></td><td class="num">101.9 <span class="delta">(-32.813)</span></td><td class="num">147.4 <span class="delta">(-26.020)</span></td><td class="num">77.9 <span class="delta">(-0.095)</span></td><td class="num">79.1 <span class="delta">(-0.493)</span></td><td class="num">28871 <span class="delta">(-786.728)</span></td></tr><tr class="baseline-row"><td>resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.7 <span class="delta"></span></td><td class="num">134.7 <span class="delta"></span></td><td class="num">173.4 <span class="delta"></span></td><td class="num">78.0 <span class="delta"></span></td><td class="num">79.6 <span class="delta"></span></td><td class="num">29658 <span class="delta"></span></td></tr><tr><td>resnext50_32x4d_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.4 <span class="delta">(-0.363)</span></td><td class="num">381.1 <span class="delta">(+246.442)</span></td><td class="num">514.3 <span class="delta">(+340.840)</span></td><td class="num">79.5 <span class="delta">(+1.498)</span></td><td class="num">84.1 <span class="delta">(+4.491)</span></td><td class="num">45881 <span class="delta">(+16223.426)</span></td></tr><tr><td>swin_t_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">51.8 <span class="delta">(-1.992)</span></td><td class="num">481.1 <span class="delta">(+346.406)</span></td><td class="num">512.4 <span class="delta">(+338.946)</span></td><td class="num">80.2 <span class="delta">(+2.211)</span></td><td class="num">85.8 <span class="delta">(+6.269)</span></td><td class="num">45197 <span class="delta">(+15539.700)</span></td></tr><tr><td>vgg16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.1 <span class="delta">(-1.688)</span></td><td class="num">218.4 <span class="delta">(+83.705)</span></td><td class="num">287.3 <span class="delta">(+113.858)</span></td><td class="num">78.0 <span class="delta">(-0.021)</span></td><td class="num">79.3 <span class="delta">(-0.265)</span></td><td class="num">31082 <span class="delta">(+1424.408)</span></td></tr><tr><td>vit_b_16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">51.7 <span class="delta">(-2.015)</span></td><td class="num">562.2 <span class="delta">(+427.473)</span></td><td class="num">585.8 <span class="delta">(+412.344)</span></td><td class="num">80.2 <span class="delta">(+2.190)</span></td><td class="num">87.0 <span class="delta">(+7.404)</span></td><td class="num">48888 <span class="delta">(+19230.407)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_6" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["convnext_tiny_cifar10_bs64_nocap_fixed", "densenet121_cifar10_bs64_nocap_fixed", "efficientnet_b0_cifar10_bs64_nocap_fixed", "mobilenetv2_cifar10_bs64_nocap_fixed", "resnet18_cifar10_bs64_nocap_fixed", "resnet50_cifar10_bs64_nocap_fixed", "resnext50_32x4d_cifar10_bs64_nocap_fixed", "swin_t_cifar10_bs64_nocap_fixed", "vgg16_cifar10_bs64_nocap_fixed", "vit_b_16_cifar10_bs64_nocap_fixed"]; |
| var barTrace = { |
| x: labels, y: [325.6, 378.6, 310.7, 103.0, 101.9, 134.7, 381.1, 481.1, 218.4, 562.2], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [79.1, 79.2, 78.9, 78.0, 77.9, 78.0, 79.5, 80.2, 78.0, 80.2], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_6', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_6')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_6" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">convnext_tiny_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_convnext_tiny_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">densenet121_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_densenet121_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">efficientnet_b0_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_efficientnet_b0_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mobilenetv2_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_mobilenetv2_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnext50_32x4d_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_resnext50_32x4d_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">swin_t_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_swin_t_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">vgg16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_vgg16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">vit_b_16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/06_model_scaling_image/fixed_vit_b_16_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#07</span> |
| <span class="exp-title">데이터셋별 전력 비교</span> |
| </div> |
| <p class="exp-desc">데이터셋(이미지 해상도, 클래스 수)이 전력 프로파일에 미치는 영향을 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 데이터셋 (CIFAR-10 / CIFAR-100 / ImageNet)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr class="baseline-row"><td>resnet50_cifar100_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.4 <span class="delta"></span></td><td class="num">131.6 <span class="delta"></span></td><td class="num">170.0 <span class="delta"></span></td><td class="num">78.2 <span class="delta"></span></td><td class="num">79.3 <span class="delta"></span></td><td class="num">29700 <span class="delta"></span></td></tr><tr><td>resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.7 <span class="delta">(+1.284)</span></td><td class="num">135.0 <span class="delta">(+3.411)</span></td><td class="num">172.6 <span class="delta">(+2.617)</span></td><td class="num">78.3 <span class="delta">(+0.122)</span></td><td class="num">79.5 <span class="delta">(+0.127)</span></td><td class="num">29821 <span class="delta">(+121.380)</span></td></tr><tr><td>resnet50_imagenet_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">52.0 <span class="delta">(-0.356)</span></td><td class="num">479.7 <span class="delta">(+348.097)</span></td><td class="num">490.5 <span class="delta">(+320.526)</span></td><td class="num">79.0 <span class="delta">(+0.804)</span></td><td class="num">83.4 <span class="delta">(+4.041)</span></td><td class="num">338171 <span class="delta">(+308470.969)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_7" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["resnet50_cifar100_bs64_nocap_fixed", "resnet50_cifar10_bs64_nocap_fixed", "resnet50_imagenet_bs64_nocap_fixed"]; |
| var barTrace = { |
| x: labels, y: [131.6, 135.0, 479.7], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [78.2, 78.3, 79.0], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_7', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_7')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_7" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar100_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/07_dataset_cifar10_cifar100_imagenet/fixed_resnet50_cifar100_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/07_dataset_cifar10_cifar100_imagenet/fixed_resnet50_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet50_imagenet_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/07_dataset_cifar10_cifar100_imagenet/fixed_resnet50_imagenet_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#08</span> |
| <span class="exp-title">학습 모드 비교</span> |
| </div> |
| <p class="exp-desc">고정 SGD 학습 vs Optuna 기반 자동 하이퍼파라미터 탐색의 전력 비교입니다.</p> |
| <span class="exp-variable">실험 변수: 학습 모드 (fixed / automl)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">54.2 <span class="delta"></span></td><td class="num">102.5 <span class="delta"></span></td><td class="num">147.7 <span class="delta"></span></td><td class="num">77.9 <span class="delta"></span></td><td class="num">79.2 <span class="delta"></span></td><td class="num">28945 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_8')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_8" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/08_train_modes_fixed_automl/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#10</span> |
| <span class="exp-title">체크포인트 I/O 관찰</span> |
| </div> |
| <p class="exp-desc">주기적 모델 저장이 만드는 전력 스파이크(GPU 연산 중단 → 디스크 I/O)를 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 체크포인트 주기</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</td><td class="num">53.5 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">77.9 <span class="delta"></span></td><td class="num">79.0 <span class="delta"></span></td><td class="num">29034 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_9')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_9" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed</div> |
| <iframe src="gpu_power_experiment/10_checkpoint_iteration_observe/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#11</span> |
| <span class="exp-title">클럭 주파수 고정 실험</span> |
| </div> |
| <p class="exp-desc">GPU SM 클럭을 특정 주파수(1005/1500/2100MHz)로 고정했을 때의 전력 변화를 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 클럭 고정 (1005 / 1500 / 2100 MHz)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 Idle 평균 (W)</th><th>추론 피크 (W)</th><th>총 에너지 (J)</th></tr></thead> |
| <tbody><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1005MHz_fixed</td><td class="num">52.9 <span class="delta">(-8.270)</span></td><td class="num">64.4 <span class="delta">(-11.827)</span></td><td class="num">79.1 <span class="delta">(-18.937)</span></td><td class="num">53.5 <span class="delta">(-8.253)</span></td><td class="num">54.7 <span class="delta">(-7.955)</span></td><td class="num">21503 <span class="delta">(-3294.508)</span></td></tr><tr><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1500MHz_fixed</td><td class="num">57.9 <span class="delta">(-3.234)</span></td><td class="num">70.9 <span class="delta">(-5.391)</span></td><td class="num">88.1 <span class="delta">(-9.952)</span></td><td class="num">57.9 <span class="delta">(-3.928)</span></td><td class="num">58.3 <span class="delta">(-4.400)</span></td><td class="num">23260 <span class="delta">(-1537.386)</span></td></tr><tr class="baseline-row"><td>resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk2100MHz_fixed</td><td class="num">61.2 <span class="delta"></span></td><td class="num">76.3 <span class="delta"></span></td><td class="num">98.1 <span class="delta"></span></td><td class="num">61.8 <span class="delta"></span></td><td class="num">62.7 <span class="delta"></span></td><td class="num">24797 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_gpu_power_experiment_10" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["resnet18_cifar10_bs64_nocap_clk1005MH…", "resnet18_cifar10_bs64_nocap_clk1500MH…", "resnet18_cifar10_bs64_nocap_clk2100MH…"]; |
| var barTrace = { |
| x: labels, y: [64.4, 70.9, 76.3], type: 'bar', name: '학습 평균 전력 (W)', |
| marker: { color: '#818cf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [53.5, 57.9, 61.8], type: 'scatter', mode: 'lines+markers', |
| name: '추론 평균 전력 (W)', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 학습 전력 vs 추론 전력', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: '전력 (W)', titlefont: { color: '#818cf8' }, |
| tickfont: { color: '#818cf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: '전력 (W)', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_gpu_power_experiment_10', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_gpu_power_experiment_10')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_gpu_power_experiment_10" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1005MHz_fixed</div> |
| <iframe src="gpu_power_experiment/11_control_clock_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1005MHz_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1500MHz_fixed</div> |
| <iframe src="gpu_power_experiment/11_control_clock_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk1500MHz_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk2100MHz_fixed</div> |
| <iframe src="gpu_power_experiment/11_control_clock_sweep/fixed_resnet18_cifar10_gpu0_bs64_ep1_req10000_nocap_clk2100MHz_fixed/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| </div> |
|
|
| <div id="tab-llm" class="tab-content"> |
| <div class="section-divider"><h2>🤖 LLM 실험 결과</h2></div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#01</span> |
| <span class="exp-title">LLM 모델 스케일링 (초기)</span> |
| </div> |
| <p class="exp-desc">GPT-2, Qwen3-4B, Mistral-7B, Llama3.1-8B 네 종류의 LLM 전력 프로파일을 비교합니다.</p> |
| <span class="exp-variable">실험 변수: LLM 모델 종류</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>gpt2_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">53.6 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">77.3 <span class="delta"></span></td><td class="num">92.2 <span class="delta"></span></td><td class="num">377.485 <span class="delta"></span></td><td class="num">0.204 <span class="delta"></span></td></tr><tr><td>llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">53.4 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">226.5 <span class="delta"></span></td><td class="num">435.6 <span class="delta"></span></td><td class="num">76.366 <span class="delta"></span></td><td class="num">2.968 <span class="delta"></span></td></tr><tr><td>mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">54.1 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">221.7 <span class="delta"></span></td><td class="num">428.0 <span class="delta"></span></td><td class="num">79.432 <span class="delta"></span></td><td class="num">2.792 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">54.7 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">173.3 <span class="delta"></span></td><td class="num">298.2 <span class="delta"></span></td><td class="num">72.853 <span class="delta"></span></td><td class="num">2.379 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_0" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["gpt2alpaca_nocap", "llama3.1-8balpaca_nocap", "mistral-7balpaca_nocap", "qwen3-4balpaca_nocap"]; |
| var barTrace = { |
| x: labels, y: [377.49, 76.37, 79.43, 72.85], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [0.204, 2.968, 2.792, 2.379], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_0', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_0')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_0" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">gpt2_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_gpt2_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">gpt2_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_gpt2_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive_gpt2.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_llama3.1-8b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_llama3.1-8b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive_llama.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_mistral-7b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_mistral-7b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive_mistral.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/01_model_scaling_initial/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive_qwen.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#02</span> |
| <span class="exp-title">디코드 길이 비교 (32 vs 128)</span> |
| </div> |
| <p class="exp-desc">최대 토큰 생성 길이(32/128)에 따른 전력 변화를 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 최대 생성 토큰 수 (32 / 128)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">53.5 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">175.0 <span class="delta"></span></td><td class="num">307.1 <span class="delta"></span></td><td class="num">75.556 <span class="delta"></span></td><td class="num">2.315 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</td><td class="num">52.5 <span class="delta">(-1.022)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">85.2 <span class="delta">(-89.816)</span></td><td class="num">142.6 <span class="delta">(-164.580)</span></td><td class="num">76.548 <span class="delta">(+0.992)</span></td><td class="num">1.116 <span class="delta">(-1.199)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_1" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_nocap", "qwen3-4balpaca_maxtok32_nocap"]; |
| var barTrace = { |
| x: labels, y: [75.56, 76.55], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [2.315, 1.116], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_1', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_1')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_1" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/02_decode_length_32_128/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/02_decode_length_32_128/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</div> |
| <iframe src="llm_power_experiment/02_decode_length_32_128/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</div> |
| <iframe src="llm_power_experiment/02_decode_length_32_128/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#03</span> |
| <span class="exp-title">LLM 데이터셋 비교</span> |
| </div> |
| <p class="exp-desc">Alpaca, MMLU-Pro, LongBench 데이터셋별 전력 프로파일을 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 데이터셋 (alpaca / mmlu-pro / longbench)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">55.1 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">173.8 <span class="delta"></span></td><td class="num">306.1 <span class="delta"></span></td><td class="num">75.681 <span class="delta"></span></td><td class="num">2.296 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_longbench_maxtok128_nocap</td><td class="num">52.6 <span class="delta">(-2.582)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">226.6 <span class="delta">(+52.854)</span></td><td class="num">357.8 <span class="delta">(+51.734)</span></td><td class="num">70.731 <span class="delta">(-4.950)</span></td><td class="num">3.205 <span class="delta">(+0.909)</span></td></tr><tr><td>qwen3-4b_gpu0_ds_mmlu-pro_maxtok128_nocap</td><td class="num">53.0 <span class="delta">(-2.111)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">183.5 <span class="delta">(+9.768)</span></td><td class="num">294.9 <span class="delta">(-11.166)</span></td><td class="num">71.134 <span class="delta">(-4.547)</span></td><td class="num">2.581 <span class="delta">(+0.285)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_2" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_nocap", "qwen3-4blongbench_nocap", "qwen3-4bmmlu-pro_nocap"]; |
| var barTrace = { |
| x: labels, y: [75.68, 70.73, 71.13], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [2.296, 3.205, 2.581], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_2', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_2')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_2" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_longbench_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_longbench_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_longbench_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_longbench_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_longbench_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_mmlu-pro_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_mmlu-pro_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_mmlu-pro_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/03_dataset_alpaca_mmlu_longbench/llm_qwen3-4b_gpu0_ds_mmlu-pro_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_mmlu-pro_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#04</span> |
| <span class="exp-title">디코드 길이 확장 비교 (32/256/512)</span> |
| </div> |
| <p class="exp-desc">더 넓은 범위의 최대 생성 토큰 수(32/256/512)별 전력 변화를 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 최대 생성 토큰 수 (32 / 256 / 512)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok256_nocap</td><td class="num">53.4 <span class="delta">(+0.562)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">232.1 <span class="delta">(+148.068)</span></td><td class="num">296.9 <span class="delta">(+132.907)</span></td><td class="num">70.766 <span class="delta">(-2.000)</span></td><td class="num">3.280 <span class="delta">(+2.128)</span></td></tr><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</td><td class="num">52.8 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">84.0 <span class="delta"></span></td><td class="num">164.0 <span class="delta"></span></td><td class="num">72.766 <span class="delta"></span></td><td class="num">1.152 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok512_nocap</td><td class="num">52.5 <span class="delta">(-0.364)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">266.1 <span class="delta">(+182.151)</span></td><td class="num">309.2 <span class="delta">(+145.249)</span></td><td class="num">75.804 <span class="delta">(+3.038)</span></td><td class="num">3.511 <span class="delta">(+2.359)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_3" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_maxtok256_nocap", "qwen3-4balpaca_maxtok32_nocap", "qwen3-4balpaca_maxtok512_nocap"]; |
| var barTrace = { |
| x: labels, y: [70.77, 72.77, 75.8], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [3.28, 1.152, 3.511], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_3', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_3')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_3" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok256_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok256_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok256_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok256_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok256_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok32_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok32_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok512_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok512_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok512_nocap</div> |
| <iframe src="llm_power_experiment/04_decode_length_32_256_512/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok512_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok512_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#05</span> |
| <span class="exp-title">LLM 추론 패턴 비교</span> |
| </div> |
| <p class="exp-desc">고정/가변/버스트/동시 버스트 네 가지 추론 패턴이 전력과 처리량에 미치는 영향을 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 추론 패턴 (fixed / variable / burst / concurrent_burst)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap</td><td class="num">52.1 <span class="delta">(+0.794)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">214.3 <span class="delta">(+40.486)</span></td><td class="num">306.6 <span class="delta">(-0.193)</span></td><td class="num">75.905 <span class="delta">(-0.029)</span></td><td class="num">3.948 <span class="delta">(+1.659)</span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap</td><td class="num">51.7 <span class="delta">(+0.393)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">284.0 <span class="delta">(+110.274)</span></td><td class="num">302.3 <span class="delta">(-4.493)</span></td><td class="num">74.369 <span class="delta">(-1.565)</span></td><td class="num">3.819 <span class="delta">(+1.530)</span></td></tr><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">51.3 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">173.8 <span class="delta"></span></td><td class="num">306.8 <span class="delta"></span></td><td class="num">75.934 <span class="delta"></span></td><td class="num">2.289 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap</td><td class="num">51.0 <span class="delta">(-0.314)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">175.9 <span class="delta">(+2.176)</span></td><td class="num">299.1 <span class="delta">(-7.750)</span></td><td class="num">73.076 <span class="delta">(-2.858)</span></td><td class="num">2.408 <span class="delta">(+0.119)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_4" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_burst_nocap", "qwen3-4balpaca_concurrent_burst_nocap", "qwen3-4balpaca_nocap", "qwen3-4balpaca_variable_nocap"]; |
| var barTrace = { |
| x: labels, y: [75.91, 74.37, 75.93, 73.08], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [3.948, 3.819, 2.289, 2.408], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_4', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_4')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_4" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_burst_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_concurrent_burst_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/05_inference_pattern_fixed_var_burst/llm_qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_variable_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#06</span> |
| <span class="exp-title">LLM 전력 상한 (Cap) 실험</span> |
| </div> |
| <p class="exp-desc">전력 제한(200~575W)이 LLM 추론 성능(tokens/s)과 효율(J/token)에 미치는 영향을 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 전력 상한 (200W ~ 575W)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap200W</td><td class="num">52.6 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">142.6 <span class="delta"></span></td><td class="num">215.5 <span class="delta"></span></td><td class="num">71.056 <span class="delta"></span></td><td class="num">2.007 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap240W</td><td class="num">51.2 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">158.2 <span class="delta"></span></td><td class="num">255.3 <span class="delta"></span></td><td class="num">74.859 <span class="delta"></span></td><td class="num">2.114 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap280W</td><td class="num">53.3 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">169.8 <span class="delta"></span></td><td class="num">291.9 <span class="delta"></span></td><td class="num">76.121 <span class="delta"></span></td><td class="num">2.231 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap320W</td><td class="num">52.3 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">177.3 <span class="delta"></span></td><td class="num">303.0 <span class="delta"></span></td><td class="num">74.011 <span class="delta"></span></td><td class="num">2.396 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap360W</td><td class="num">52.7 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">172.0 <span class="delta"></span></td><td class="num">305.1 <span class="delta"></span></td><td class="num">75.777 <span class="delta"></span></td><td class="num">2.269 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_5" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_cap200W", "qwen3-4balpaca_cap240W", "qwen3-4balpaca_cap280W", "qwen3-4balpaca_cap320W", "qwen3-4balpaca_cap360W"]; |
| var barTrace = { |
| x: labels, y: [71.06, 74.86, 76.12, 74.01, 75.78], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [2.007, 2.114, 2.231, 2.396, 2.269], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_5', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_5')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_5" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap200W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap200W/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap240W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap240W/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap280W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap280W/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap320W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap320W/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap360W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap360W/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap360W</div> |
| <iframe src="llm_power_experiment/06_power_cap_200_240_280_320_360/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap360W/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap360W__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#07</span> |
| <span class="exp-title">LLM 클럭 고정 실험</span> |
| </div> |
| <p class="exp-desc">GPU 클럭 고정(1005/1500/2100MHz)이 LLM 추론 성능에 미치는 영향을 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 클럭 고정 (1005 / 1500 / 2100 MHz)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1005MHz</td><td class="num">52.6 <span class="delta">(-8.685)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">107.9 <span class="delta">(-20.127)</span></td><td class="num">146.1 <span class="delta">(-53.238)</span></td><td class="num">55.456 <span class="delta">(-16.889)</span></td><td class="num">1.949 <span class="delta">(+0.179)</span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1500MHz</td><td class="num">57.4 <span class="delta">(-3.855)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">119.2 <span class="delta">(-8.922)</span></td><td class="num">176.6 <span class="delta">(-22.724)</span></td><td class="num">68.873 <span class="delta">(-3.472)</span></td><td class="num">1.730 <span class="delta">(-0.040)</span></td></tr><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk2100MHz</td><td class="num">61.3 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">128.1 <span class="delta"></span></td><td class="num">199.3 <span class="delta"></span></td><td class="num">72.345 <span class="delta"></span></td><td class="num">1.770 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_6" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_nocap_clk1005MHz", "qwen3-4balpaca_nocap_clk1500MHz", "qwen3-4balpaca_nocap_clk2100MHz"]; |
| var barTrace = { |
| x: labels, y: [55.46, 68.87, 72.34], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [1.949, 1.73, 1.77], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_6', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_6')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_6" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1005MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1005MHz/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1005MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1005MHz/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1005MHz__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1500MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1500MHz/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk1500MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1500MHz/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk1500MHz__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk2100MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk2100MHz/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_clk2100MHz</div> |
| <iframe src="llm_power_experiment/07_clock_lock_1005_1500_2100/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk2100MHz/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_clk2100MHz__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#08</span> |
| <span class="exp-title">LLM 전력 램프 실험</span> |
| </div> |
| <p class="exp-desc">단계적 전력 변화가 LLM 추론 중 전력 소비에 미치는 영향을 관찰합니다.</p> |
| <span class="exp-variable">실험 변수: 램프 활성화</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_ramp</td><td class="num">53.0 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">172.2 <span class="delta"></span></td><td class="num">303.0 <span class="delta"></span></td><td class="num">75.680 <span class="delta"></span></td><td class="num">2.276 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_7')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_7" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_ramp</div> |
| <iframe src="llm_power_experiment/08_ramp_rate/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_ramp/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_ramp</div> |
| <iframe src="llm_power_experiment/08_ramp_rate/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_ramp/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_ramp__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#09</span> |
| <span class="exp-title">복합 제어 실험 (Cap + Clock + Ramp)</span> |
| </div> |
| <p class="exp-desc">전력 상한, 클럭 고정, 램프를 조합한 복합 제어의 효과를 테스트합니다.</p> |
| <span class="exp-variable">실험 변수: 제어 조합</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_clk1500MHz</td><td class="num">57.8 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">114.6 <span class="delta"></span></td><td class="num">176.2 <span class="delta"></span></td><td class="num">68.174 <span class="delta"></span></td><td class="num">1.682 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_ramp</td><td class="num">52.6 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">176.7 <span class="delta"></span></td><td class="num">296.4 <span class="delta"></span></td><td class="num">71.618 <span class="delta"></span></td><td class="num">2.466 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_8" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_cap345W_clk1500MHz", "qwen3-4balpaca_cap345W_ramp"]; |
| var barTrace = { |
| x: labels, y: [68.17, 71.62], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [1.682, 2.466], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_8', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_8')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_8" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_clk1500MHz</div> |
| <iframe src="llm_power_experiment/09_control_combo_cap_clock_ramp/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_clk1500MHz/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_clk1500MHz</div> |
| <iframe src="llm_power_experiment/09_control_combo_cap_clock_ramp/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_clk1500MHz/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_clk1500MHz__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_ramp</div> |
| <iframe src="llm_power_experiment/09_control_combo_cap_clock_ramp/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_ramp/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_cap345W_ramp</div> |
| <iframe src="llm_power_experiment/09_control_combo_cap_clock_ramp/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_ramp/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_cap345W_ramp__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#10</span> |
| <span class="exp-title">정밀도별 전력 비교 (BF16/FP16/4bit)</span> |
| </div> |
| <p class="exp-desc">연산 정밀도(BF16/FP16/4bit 양자화)에 따른 전력 및 성능 변화를 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 정밀도 (bf16 / fp16 / 4bit)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr class="baseline-row"><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">50.8 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">182.0 <span class="delta"></span></td><td class="num">302.7 <span class="delta"></span></td><td class="num">73.794 <span class="delta"></span></td><td class="num">2.467 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_4bit</td><td class="num">52.6 <span class="delta">(+1.808)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">139.8 <span class="delta">(-42.113)</span></td><td class="num">168.8 <span class="delta">(-133.883)</span></td><td class="num">40.356 <span class="delta">(-33.438)</span></td><td class="num">3.464 <span class="delta">(+0.997)</span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_fp16</td><td class="num">52.4 <span class="delta">(+1.675)</span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">180.0 <span class="delta">(-1.960)</span></td><td class="num">295.7 <span class="delta">(-7.002)</span></td><td class="num">71.605 <span class="delta">(-2.189)</span></td><td class="num">2.514 <span class="delta">(+0.047)</span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_9" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["qwen3-4balpaca_nocap", "qwen3-4balpaca_nocap_4bit", "qwen3-4balpaca_nocap_fp16"]; |
| var barTrace = { |
| x: labels, y: [73.79, 40.36, 71.61], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [2.467, 3.464, 2.514], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_9', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_9')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_9" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_4bit</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_4bit/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_4bit</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_4bit/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_4bit__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_fp16</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_fp16/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap_fp16</div> |
| <iframe src="llm_power_experiment/10_precision_bf16_fp16_4bit/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_fp16/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap_fp16__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#11</span> |
| <span class="exp-title">학습/추론 분리 실험</span> |
| </div> |
| <p class="exp-desc">학습만, 추론만, 전체 사이클의 전력 프로파일 차이를 비교합니다.</p> |
| <span class="exp-variable">실험 변수: 실행 구간 (train_only / infer_only / full_cycle)</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>mistral-7b_gpu0_ds_longbench_maxtok256_nocap</td><td class="num">53.9 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">308.6 <span class="delta"></span></td><td class="num">409.1 <span class="delta"></span></td><td class="num">49.623 <span class="delta"></span></td><td class="num">6.217 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">51.9 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">178.0 <span class="delta"></span></td><td class="num">304.0 <span class="delta"></span></td><td class="num">71.128 <span class="delta"></span></td><td class="num">2.502 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">55.5 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_10" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["mistral-7blongbench_maxtok256_nocap", "qwen3-4balpaca_nocap", "qwen3-4balpaca_nocap"]; |
| var barTrace = { |
| x: labels, y: [49.62, 71.13, 0], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [6.217, 2.502, 0], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_10', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_10')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_10" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_longbench_maxtok256_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_mistral-7b_gpu0_ds_longbench_fixed_maxtok256_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_longbench_maxtok256_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_mistral-7b_gpu0_ds_longbench_fixed_maxtok256_nocap/plots/llm_mistral-7b_gpu0_ds_longbench_fixed_maxtok256_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/11_train_infer_split/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| <div class="exp-card"> |
| <div class="exp-header"> |
| <span class="exp-number">#12</span> |
| <span class="exp-title">LLM 모델별 토큰-전력 벤치마크</span> |
| </div> |
| <p class="exp-desc">GPT-2, Qwen3-4B, Mistral-7B, Llama3.1-8B의 tokens/s 및 J/token을 벤치마크합니다.</p> |
| <span class="exp-variable">실험 변수: LLM 모델 종류</span> |
| |
| <table class="comp-table"> |
| <thead><tr><th>조건</th><th>Idle 평균 (W)</th><th>학습 평균 (W)</th><th>학습 피크 (W)</th><th>추론 평균 (W)</th><th>추론 피크 (W)</th><th>Decode Token/s (↑)</th><th>J/Token (↓)</th></tr></thead> |
| <tbody><tr><td>gpt2_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">54.6 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">77.5 <span class="delta"></span></td><td class="num">97.3 <span class="delta"></span></td><td class="num">386.545 <span class="delta"></span></td><td class="num">0.201 <span class="delta"></span></td></tr><tr><td>llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">54.0 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">245.7 <span class="delta"></span></td><td class="num">435.0 <span class="delta"></span></td><td class="num">76.413 <span class="delta"></span></td><td class="num">3.214 <span class="delta"></span></td></tr><tr><td>mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">51.3 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">222.3 <span class="delta"></span></td><td class="num">426.9 <span class="delta"></span></td><td class="num">79.479 <span class="delta"></span></td><td class="num">2.797 <span class="delta"></span></td></tr><tr><td>qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</td><td class="num">52.2 <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">— <span class="delta"></span></td><td class="num">173.7 <span class="delta"></span></td><td class="num">302.9 <span class="delta"></span></td><td class="num">73.947 <span class="delta"></span></td><td class="num">2.349 <span class="delta"></span></td></tr></tbody> |
| </table> |
| |
| |
| <div class="comparison-chart-wrapper"> |
| <div id="chart_llm_power_experiment_11" class="comparison-chart"></div> |
| <script> |
| (function() { |
| var labels = ["gpt2alpaca_nocap", "llama3.1-8balpaca_nocap", "mistral-7balpaca_nocap", "qwen3-4balpaca_nocap"]; |
| var barTrace = { |
| x: labels, y: [386.55, 76.41, 79.48, 73.95], type: 'bar', name: 'Decode Token/s', |
| marker: { color: '#38bdf8', opacity: 0.85 }, |
| yaxis: 'y' |
| }; |
| var lineTrace = { |
| x: labels, y: [0.201, 3.214, 2.797, 2.349], type: 'scatter', mode: 'lines+markers', |
| name: 'J/Token', line: { color: '#f87171', width: 2.5 }, |
| marker: { size: 7 }, yaxis: 'y2' |
| }; |
| var layout = { |
| title: { text: '조건별 처리량 vs 에너지 효율', font: { size: 14, color: '#e2e8f0' } }, |
| paper_bgcolor: '#162032', plot_bgcolor: '#0f172a', |
| font: { color: '#94a3b8', size: 11 }, |
| xaxis: { tickangle: -40, tickfont: { size: 9 }, gridcolor: '#1e293b' }, |
| yaxis: { title: 'Token/s', titlefont: { color: '#38bdf8' }, |
| tickfont: { color: '#38bdf8' }, gridcolor: '#1e293b', side: 'left' }, |
| yaxis2: { title: 'J/Token', titlefont: { color: '#f87171' }, |
| tickfont: { color: '#f87171' }, overlaying: 'y', |
| side: 'right', gridcolor: 'transparent' }, |
| legend: { orientation: 'h', y: 1.12, x: 0.5, xanchor: 'center', |
| font: { size: 11 } }, |
| margin: { l: 60, r: 60, t: 60, b: 130 }, |
| bargap: 0.3 |
| }; |
| Plotly.newPlot('chart_llm_power_experiment_11', [barTrace, lineTrace], layout, |
| { responsive: true, displayModeBar: false }); |
| })(); |
| </script> |
| </div> |
| |
| <div class="collapsible-toggle" onclick="toggleSection('graphs_llm_power_experiment_11')">▶ 개별 전력 그래프 펼치기</div> |
| <div class="collapsible-content" id="graphs_llm_power_experiment_11" style="display:none;"> |
| <div class="graph-grid"> |
| <div class="graph-item"> |
| <div class="graph-label">gpt2_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_gpt2_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">gpt2_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_gpt2_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_gpt2_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_llama3.1-8b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">llama3.1-8b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_llama3.1-8b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_llama3.1-8b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_mistral-7b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">mistral-7b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_mistral-7b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_mistral-7b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| |
| <div class="graph-item"> |
| <div class="graph-label">qwen3-4b_gpu0_ds_alpaca_maxtok128_nocap</div> |
| <iframe src="llm_power_experiment/12_model_scaling_tokenpowerbench/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap/plots/llm_qwen3-4b_gpu0_ds_alpaca_fixed_maxtok128_nocap__gpu_power_interactive.html?v=1771485347" loading="lazy"></iframe> |
| </div> |
| </div> |
| </div> |
| </div> |
| |
| </div> |
|
|
| <div class="footer"> |
| <p>GPU/LLM 전력 실험 전송 패키지 · 2026-02-19 16:15 생성</p> |
| <p>각 run 폴더 구성: config.json · phase_power_summary.csv · gpu_samples.csv · gpu_power_interactive.html · gpu_metrics.png</p> |
| </div> |
| </div> |
| <script> |
| function switchTab(tabId) { |
| document.querySelectorAll('.tab-btn').forEach(b => b.classList.remove('active')); |
| document.querySelectorAll('.tab-content').forEach(c => c.classList.remove('active')); |
| document.getElementById('tab-' + tabId).classList.add('active'); |
| document.querySelector('[onclick*="' + tabId + '"]').classList.add('active'); |
| } |
| function toggleSection(id) { |
| var el = document.getElementById(id); |
| var toggle = el.previousElementSibling; |
| if (el.style.display === 'none') { |
| el.style.display = 'block'; |
| toggle.textContent = '▼ 그래프 접기'; |
| } else { |
| el.style.display = 'none'; |
| toggle.textContent = '▶ 그래프 펼치기'; |
| } |
| } |
| </script> |
| </body> |
| </html> |
