Split bee-bench into perf and power workflows

audit/internal/webui/pages.go

@@ -1946,7 +1946,7 @@ func renderBenchmark(opts HandlerOptions) string {
 
 <div class="grid2">
   <div class="card">
-    <div class="card-head">NVIDIA Benchmark</div>
+    <div class="card-head">Benchmark Setup</div>
     <div class="card-body">
       <div class="form-row">
         <label>Profile</label>
@@ -1979,21 +1979,25 @@ func renderBenchmark(opts HandlerOptions) string {
         <span>Ramp-up — 1 GPU → 2 → … → all selected (separate tasks)</span>
       </label>
       <p id="benchmark-selection-note" style="font-size:12px;color:var(--muted);margin:10px 0 14px">Select one GPU for single-card benchmarking or several GPUs for a constrained multi-GPU run.</p>
-      <button id="benchmark-run-btn" class="btn btn-primary" onclick="runNvidiaBenchmark()" disabled>&#9654; Run Benchmark</button>
+      <div style="display:flex;gap:8px;flex-wrap:wrap;align-items:center">
+        <button id="benchmark-run-performance-btn" class="btn btn-primary" onclick="runNvidiaBenchmark('performance')" disabled>&#9654; Run Performance Benchmark</button>
+        <button id="benchmark-run-power-fit-btn" class="btn btn-secondary" onclick="runNvidiaBenchmark('power-fit')" disabled>&#9654; Run Power / Thermal Fit</button>
+      </div>
+      <span id="benchmark-run-nccl" hidden>nccl-auto</span>
       <span id="benchmark-run-status" style="margin-left:10px;font-size:12px;color:var(--muted)"></span>
     </div>
   </div>
 
   <div class="card">
-    <div class="card-head">Method</div>
+    <div class="card-head">Method Split</div>
     <div class="card-body">
-      <p style="font-size:13px;color:var(--muted);margin-bottom:10px">Each benchmark run performs warmup, sustained compute, telemetry capture, cooldown, and optional NCCL interconnect checks.</p>
+      <p style="font-size:13px;color:var(--muted);margin-bottom:10px">The benchmark page now exposes two fundamentally different test families so compute score and server power-fit are not mixed into one number.</p>
       <table>
-        <tr><th>Profile</th><th>Purpose</th></tr>
-        <tr><td>Standard</td><td>Fast, repeatable performance check for server-to-server comparison.</td></tr>
-        <tr><td>Stability</td><td>Longer run for thermal drift, power caps, and clock instability.</td></tr>
-        <tr><td>Overnight</td><td>Extended verification of long-run stability and late throttling.</td></tr>
+        <tr><th>Run Type</th><th>Engine</th><th>Question</th></tr>
+        <tr><td>Performance Benchmark</td><td><code>bee-gpu-burn</code></td><td>How much isolated compute performance does the GPU realize in this server?</td></tr>
+        <tr><td>Power / Thermal Fit</td><td><code>dcgmi targeted_power</code></td><td>How much power per GPU can this server sustain as GPU count ramps up?</td></tr>
       </table>
+      <p style="font-size:12px;color:var(--muted);margin-top:10px">Use ramp-up mode for capacity work: it creates 1 GPU → 2 GPU → … → all selected steps so analysis software can derive server total score and watts-per-GPU curves.</p>
     </div>
   </div>
 </div>
@@ -2036,21 +2040,24 @@ function benchmarkMode() {
 
 function benchmarkUpdateSelectionNote() {
   const selected = benchmarkSelectedGPUIndices();
-  const btn = document.getElementById('benchmark-run-btn');
+  const perfBtn = document.getElementById('benchmark-run-performance-btn');
+  const fitBtn = document.getElementById('benchmark-run-power-fit-btn');
   const note = document.getElementById('benchmark-selection-note');
   if (!selected.length) {
-    btn.disabled = true;
+    perfBtn.disabled = true;
+    fitBtn.disabled = true;
     note.textContent = 'Select at least one NVIDIA GPU to run the benchmark.';
     return;
   }
-  btn.disabled = false;
+  perfBtn.disabled = false;
+  fitBtn.disabled = false;
   const mode = benchmarkMode();
   if (mode === 'ramp-up') {
-    note.textContent = 'Ramp-up: ' + selected.length + ' tasks (1 GPU → ' + selected.length + ' GPUs). NCCL on final step.';
+    note.textContent = 'Ramp-up: ' + selected.length + ' tasks (1 GPU → ' + selected.length + ' GPUs). Performance uses compute benchmark; Power / Thermal Fit uses targeted_power per step.';
   } else if (mode === 'parallel') {
-    note.textContent = 'Parallel: all ' + selected.length + ' GPU(s) simultaneously.' + (selected.length > 1 ? ' NCCL included.' : '');
+    note.textContent = 'Parallel: all ' + selected.length + ' GPU(s) simultaneously. Only the performance benchmark supports this mode.';
   } else {
-    note.textContent = 'Sequential: each GPU benchmarked separately.' + (selected.length > 1 ? ' NCCL included on each.' : '');
+    note.textContent = 'Sequential: each selected GPU benchmarked separately.';
   }
 }
 
@@ -2124,7 +2131,7 @@ function benchmarkSelectNone() {
   benchmarkUpdateSelectionNote();
 }
 
-function runNvidiaBenchmark() {
+function runNvidiaBenchmark(kind) {
   const selected = benchmarkSelectedGPUIndices();
   const status = document.getElementById('benchmark-run-status');
   if (!selected.length) {
@@ -2134,21 +2141,26 @@ function runNvidiaBenchmark() {
   if (benchmarkES) { benchmarkES.close(); benchmarkES = null; }
   const mode = benchmarkMode();
   const rampUp = mode === 'ramp-up' && selected.length > 1;
-  const parallelGPUs = mode === 'parallel';
+  const parallelGPUs = mode === 'parallel' && kind === 'performance';
+  if (kind === 'power-fit' && mode === 'parallel') {
+    status.textContent = 'Power / Thermal Fit supports sequential or ramp-up only.';
+    return;
+  }
   const body = {
     profile: document.getElementById('benchmark-profile').value || 'standard',
     gpu_indices: selected,
-    run_nccl: selected.length > 1,
+    run_nccl: kind === 'performance' && selected.length > 1,
     parallel_gpus: parallelGPUs,
     ramp_up: rampUp,
-    display_name: 'NVIDIA Benchmark'
+    display_name: kind === 'power-fit' ? 'NVIDIA Power / Thermal Fit' : 'NVIDIA Performance Benchmark'
   };
   document.getElementById('benchmark-output').style.display = 'block';
-  document.getElementById('benchmark-title').textContent = '— ' + body.profile + ' [' + selected.join(', ') + ']';
+  document.getElementById('benchmark-title').textContent = '— ' + body.display_name + ' · ' + body.profile + ' [' + selected.join(', ') + ']';
   const term = document.getElementById('benchmark-terminal');
-  term.textContent = 'Enqueuing benchmark for GPUs ' + selected.join(', ') + '...\n';
+  term.textContent = 'Enqueuing ' + body.display_name + ' for GPUs ' + selected.join(', ') + '...\n';
   status.textContent = 'Queueing...';
-  fetch('/api/benchmark/nvidia/run', {
+  const endpoint = kind === 'power-fit' ? '/api/bee-bench/nvidia/power/run' : '/api/bee-bench/nvidia/perf/run';
+  fetch(endpoint, {
     method: 'POST',
     headers: {'Content-Type':'application/json'},
     body: JSON.stringify(body)
@@ -2202,7 +2214,7 @@ benchmarkLoadGPUs();
 func renderBenchmarkResultsCard(exportDir string) string {
 	maxIdx, runs := loadBenchmarkHistory(exportDir)
 	return renderBenchmarkResultsCardFromRuns(
-		"Benchmark Results",
+		"Perf Results",
 		"Composite score by saved benchmark run and GPU.",
 		"No saved benchmark runs yet.",
 		maxIdx,
@@ -2244,11 +2256,11 @@ func renderBenchmarkResultsCardFromRuns(title, description, emptyMessage string,
 }
 
 func loadBenchmarkHistory(exportDir string) (int, []benchmarkHistoryRun) {
-	baseDir := app.DefaultBenchmarkBaseDir
+	baseDir := app.DefaultBeeBenchPerfDir
 	if strings.TrimSpace(exportDir) != "" {
-		baseDir = filepath.Join(exportDir, "bee-benchmark")
+		baseDir = filepath.Join(exportDir, "bee-bench", "perf")
 	}
-	paths, err := filepath.Glob(filepath.Join(baseDir, "gpu-benchmark-*", "result.json"))
+	paths, err := filepath.Glob(filepath.Join(baseDir, "perf-*", "result.json"))
 	if err != nil || len(paths) == 0 {
 		return -1, nil
 	}