Capture per-run IPMI power and GPU telemetry in power benchmark

- Sample IPMI loaded_w per single-card calibration and per ramp step
  instead of averaging over the entire Phase 2; top-level ServerPower
  uses the final (all-GPU) ramp step value
- Add ServerLoadedW/ServerDeltaW to NvidiaPowerBenchGPU and
  NvidiaPowerBenchStep so external tooling can compare wall power per
  phase without re-parsing logs
- Write gpu-metrics.csv/.html inside each single-XX/ and step-XX/
  subdir; aggregate all phases into a top-level gpu-metrics.csv/.html
- Write 00-nvidia-smi-q.log at the start of every power run
- Add Telemetry (p95 temp/power/fan/clock) to NvidiaPowerBenchGPU in
  result.json from the converged calibration attempt
- Power benchmark page: split "Achieved W" into Single-card W and
  Multi-GPU W (StablePowerLimitW); derate highlight and status color
  now reflect the final multi-GPU limit vs nominal
- Performance benchmark page: add Status column and per-GPU score
  color coding (green/yellow/red) based on gpu.Status and OverallStatus

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Mikhail Chusavitin
2026-04-17 17:59:58 +03:00
parent dca4afb8d0
commit 5285c0d101
3 changed files with 178 additions and 51 deletions

View File

@@ -2014,9 +2014,11 @@ func renderSATCard(id, label, runAction, headerActions, body string) string {
// ── Benchmark ─────────────────────────────────────────────────────────────────
type benchmarkHistoryRun struct {
generatedAt time.Time
displayTime string
gpuScores map[int]float64 // GPU index → composite score
generatedAt time.Time
displayTime string
gpuScores map[int]float64 // GPU index → composite score
gpuStatuses map[int]string // GPU index → status ("OK", "WARNING", "FAILED", …)
overallStatus string
}
func renderBenchmark(opts HandlerOptions) string {
@@ -2324,7 +2326,7 @@ func renderBenchmarkResultsCardFromRuns(title, description, emptyMessage string,
b.WriteString(`<p style="color:var(--muted);font-size:13px;margin-bottom:12px">` + html.EscapeString(description) + `</p>`)
}
b.WriteString(`<div style="overflow-x:auto">`)
b.WriteString(`<table><thead><tr><th>Run</th><th>Time</th>`)
b.WriteString(`<table><thead><tr><th>Run</th><th>Time</th><th>Status</th>`)
for i := 0; i <= maxGPUIndex; i++ {
b.WriteString(`<th>GPU ` + strconv.Itoa(i) + `</th>`)
}
@@ -2333,13 +2335,36 @@ func renderBenchmarkResultsCardFromRuns(title, description, emptyMessage string,
b.WriteString(`<tr>`)
b.WriteString(`<td>#` + strconv.Itoa(i+1) + `</td>`)
b.WriteString(`<td>` + html.EscapeString(run.displayTime) + `</td>`)
overallColor := "var(--ok)"
overallLabel := run.overallStatus
if overallLabel == "" {
overallLabel = "OK"
}
if overallLabel == "FAILED" {
overallColor = "var(--crit-fg,#9f3a38)"
} else if overallLabel != "OK" {
overallColor = "var(--warn)"
}
b.WriteString(`<td style="color:` + overallColor + `;font-weight:600">` + html.EscapeString(overallLabel) + `</td>`)
for idx := 0; idx <= maxGPUIndex; idx++ {
score, ok := run.gpuScores[idx]
if !ok {
b.WriteString(`<td style="color:var(--muted)">-</td>`)
continue
}
b.WriteString(`<td>` + fmt.Sprintf("%.2f", score) + `</td>`)
gpuStatus := run.gpuStatuses[idx]
scoreColor := ""
switch gpuStatus {
case "FAILED":
scoreColor = ` style="color:var(--crit-fg,#9f3a38);font-weight:600"`
case "WARNING", "PARTIAL":
scoreColor = ` style="color:var(--warn);font-weight:600"`
case "", "OK":
// no override
default:
scoreColor = ` style="color:var(--warn);font-weight:600"`
}
b.WriteString(`<td` + scoreColor + `>` + fmt.Sprintf("%.2f", score) + `</td>`)
}
b.WriteString(`</tr>`)
}
@@ -2373,12 +2398,15 @@ func loadBenchmarkHistoryFromPaths(paths []string) (int, []benchmarkHistoryRun)
continue
}
run := benchmarkHistoryRun{
generatedAt: result.GeneratedAt,
displayTime: result.GeneratedAt.Local().Format("2006-01-02 15:04:05"),
gpuScores: make(map[int]float64),
generatedAt: result.GeneratedAt,
displayTime: result.GeneratedAt.Local().Format("2006-01-02 15:04:05"),
gpuScores: make(map[int]float64),
gpuStatuses: make(map[int]string),
overallStatus: result.OverallStatus,
}
for _, gpu := range result.GPUs {
run.gpuScores[gpu.Index] = gpu.Scores.CompositeScore
run.gpuStatuses[gpu.Index] = gpu.Status
if gpu.Index > maxGPUIndex {
maxGPUIndex = gpu.Index
}
@@ -2447,31 +2475,45 @@ func renderPowerBenchmarkResultsCard(exportDir string) string {
if len(latest.GPUs) > 0 {
b.WriteString(`<div style="overflow-x:auto"><table><thead><tr>`)
b.WriteString(`<th>GPU</th><th>Model</th><th>Nominal W</th><th>Achieved W</th><th>P95 Observed W</th><th>Status</th>`)
b.WriteString(`<th>GPU</th><th>Model</th><th>Nominal W</th><th>Single-card W</th><th>Multi-GPU W</th><th>P95 Observed W</th><th>Status</th>`)
b.WriteString(`</tr></thead><tbody>`)
for _, gpu := range latest.GPUs {
derated := gpu.Derated || (gpu.DefaultPowerLimitW > 0 && gpu.AppliedPowerLimitW < gpu.DefaultPowerLimitW-1)
// finalLimitW is the definitive TDP: multi-GPU stable limit from the ramp,
// falling back to single-card applied limit if the ramp hasn't run.
finalLimitW := gpu.StablePowerLimitW
if finalLimitW <= 0 {
finalLimitW = gpu.AppliedPowerLimitW
}
// Derate is relative to nominal (DefaultPowerLimitW), using the final limit.
derated := gpu.Derated ||
(gpu.DefaultPowerLimitW > 0 && finalLimitW > 0 && finalLimitW < gpu.DefaultPowerLimitW-1)
rowStyle := ""
achievedStyle := ""
finalStyle := ""
if derated {
rowStyle = ` style="background:rgba(255,180,0,0.08)"`
achievedStyle = ` style="color:#e6a000;font-weight:600"`
finalStyle = ` style="color:#e6a000;font-weight:600"`
}
statusLabel := gpu.Status
if statusLabel == "" {
statusLabel = "OK"
}
statusColor := "var(--ok)"
if statusLabel != "OK" {
if statusLabel == "FAILED" {
statusColor = "var(--crit-fg,#9f3a38)"
} else if statusLabel != "OK" {
statusColor = "var(--warn)"
}
nominalStr := "-"
if gpu.DefaultPowerLimitW > 0 {
nominalStr = fmt.Sprintf("%.0f", gpu.DefaultPowerLimitW)
}
achievedStr := "-"
singleStr := "-"
if gpu.AppliedPowerLimitW > 0 {
achievedStr = fmt.Sprintf("%.0f", gpu.AppliedPowerLimitW)
singleStr = fmt.Sprintf("%.0f", gpu.AppliedPowerLimitW)
}
multiStr := "-"
if gpu.StablePowerLimitW > 0 {
multiStr = fmt.Sprintf("%.0f", gpu.StablePowerLimitW)
}
p95Str := "-"
if gpu.MaxObservedPowerW > 0 {
@@ -2481,7 +2523,8 @@ func renderPowerBenchmarkResultsCard(exportDir string) string {
b.WriteString(`<td>` + strconv.Itoa(gpu.Index) + `</td>`)
b.WriteString(`<td>` + html.EscapeString(gpu.Name) + `</td>`)
b.WriteString(`<td>` + nominalStr + `</td>`)
b.WriteString(`<td` + achievedStyle + `>` + achievedStr + `</td>`)
b.WriteString(`<td>` + singleStr + `</td>`)
b.WriteString(`<td` + finalStyle + `>` + multiStr + `</td>`)
b.WriteString(`<td>` + p95Str + `</td>`)
b.WriteString(`<td style="color:` + statusColor + `;font-weight:600">` + html.EscapeString(statusLabel) + `</td>`)
b.WriteString(`</tr>`)