Add HPC tuning: PCIe ASPM off, C-states, performance CPU governor

- grub.cfg + isolinux/live.cfg.in: add pcie_aspm=off, intel_idle.max_cstate=1 and processor.max_cstate=1 to all non-failsafe boot entries - bee-hpc-tuning: new script that sets all CPU cores to performance governor via sysfs and logs THP state at boot - bee-hpc-tuning.service: runs before bee-nvidia and bee-audit - 9000-bee-setup.hook.chroot: enable service and mark script executable Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Warn on PCIe link speed degradation and collect lspci -vvv in techdump
2026-04-12 13:07:32 +03:00 · 2026-04-12 12:42:17 +03:00 · 2026-04-11 10:05:27 +03:00 · 2026-04-11 10:05:27 +03:00 · 2026-04-10 13:57:26 +03:00 · 2026-04-10 13:30:32 +03:00
36 changed files with 1276 additions and 249 deletions
--- a/audit/internal/app/app.go
+++ b/audit/internal/app/app.go
@@ -117,7 +117,7 @@ type satRunner interface {
 	RunNvidiaAcceptancePackWithOptions(ctx context.Context, baseDir string, diagLevel int, gpuIndices []int, logFunc func(string)) (string, error)
 	RunNvidiaTargetedStressValidatePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error)
 	RunNvidiaBenchmark(ctx context.Context, baseDir string, opts platform.NvidiaBenchmarkOptions, logFunc func(string)) (string, error)
-	RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error)
+	RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, staggerSec int, logFunc func(string)) (string, error)
 	RunNvidiaTargetedPowerPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error)
 	RunNvidiaPulseTestPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error)
 	RunNvidiaBandwidthPack(ctx context.Context, baseDir string, gpuIndices []int, logFunc func(string)) (string, error)
@@ -190,6 +190,7 @@ func (a *App) RunAudit(runtimeMode runtimeenv.Mode, output string) (string, erro
 	}
 	result := collector.Run(runtimeMode)
 	applyLatestSATStatuses(&result.Hardware, DefaultSATBaseDir, a.StatusDB)
+	writePSUStatusesToDB(a.StatusDB, result.Hardware.PowerSupplies)
 	if health, err := ReadRuntimeHealth(DefaultRuntimeJSONPath); err == nil {
 		result.Runtime = &health
 	}
@@ -566,11 +567,11 @@ func (a *App) RunNvidiaBenchmarkCtx(ctx context.Context, baseDir string, opts pl
 	return a.sat.RunNvidiaBenchmark(ctx, baseDir, opts, logFunc)
 }

-func (a *App) RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+func (a *App) RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, staggerSec int, logFunc func(string)) (string, error) {
 	if strings.TrimSpace(baseDir) == "" {
 		baseDir = DefaultSATBaseDir
 	}
-	return a.sat.RunNvidiaOfficialComputePack(ctx, baseDir, durationSec, gpuIndices, logFunc)
+	return a.sat.RunNvidiaOfficialComputePack(ctx, baseDir, durationSec, gpuIndices, staggerSec, logFunc)
 }

 func (a *App) RunNvidiaTargetedPowerPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
@@ -926,6 +927,41 @@ func bodyOr(body, fallback string) string {
 	return body
 }

+// writePSUStatusesToDB records PSU statuses collected during audit into the
+// component-status DB so they are visible in the Hardware Summary card.
+// PSU status is sourced from IPMI (ipmitool fru + sdr) during audit.
+func writePSUStatusesToDB(db *ComponentStatusDB, psus []schema.HardwarePowerSupply) {
+	if db == nil || len(psus) == 0 {
+		return
+	}
+	const source = "audit:ipmi"
+	worstStatus := "OK"
+	for _, psu := range psus {
+		if psu.Status == nil {
+			continue
+		}
+		slot := "?"
+		if psu.Slot != nil {
+			slot = *psu.Slot
+		}
+		st := *psu.Status
+		detail := ""
+		if psu.ErrorDescription != nil {
+			detail = *psu.ErrorDescription
+		}
+		db.Record("psu:"+slot, source, st, detail)
+		switch st {
+		case "Critical":
+			worstStatus = "Critical"
+		case "Warning":
+			if worstStatus != "Critical" {
+				worstStatus = "Warning"
+			}
+		}
+	}
+	db.Record("psu:all", source, worstStatus, "")
+}
+
 func ReadRuntimeHealth(path string) (schema.RuntimeHealth, error) {
 	raw, err := os.ReadFile(path)
 	if err != nil {
--- a/audit/internal/app/app_test.go
+++ b/audit/internal/app/app_test.go
@@ -161,7 +161,7 @@ func (f fakeSAT) RunNvidiaTargetedStressValidatePack(_ context.Context, baseDir
 	return f.runNvidiaFn(baseDir)
 }

-func (f fakeSAT) RunNvidiaOfficialComputePack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ func(string)) (string, error) {
+func (f fakeSAT) RunNvidiaOfficialComputePack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ int, _ func(string)) (string, error) {
 	if f.runNvidiaComputeFn != nil {
 		return f.runNvidiaComputeFn(baseDir, durationSec, gpuIndices)
 	}
@@ -542,8 +542,6 @@ func TestActionResultsUseFallbackBody(t *testing.T) {
 }

 func TestExportSupportBundleResultMentionsUnmountedUSB(t *testing.T) {
-	t.Parallel()
-
 	tmp := t.TempDir()
 	oldExportDir := DefaultExportDir
 	DefaultExportDir = tmp
@@ -580,8 +578,6 @@ func TestExportSupportBundleResultMentionsUnmountedUSB(t *testing.T) {
 }

 func TestExportSupportBundleResultDoesNotPretendSuccessOnError(t *testing.T) {
-	t.Parallel()
-
 	tmp := t.TempDir()
 	oldExportDir := DefaultExportDir
 	DefaultExportDir = tmp
@@ -643,8 +639,6 @@ func TestRunNvidiaAcceptancePackResult(t *testing.T) {
 }

 func TestRunSATDefaultsToExportDir(t *testing.T) {
-	t.Parallel()
-
 	oldSATBaseDir := DefaultSATBaseDir
 	DefaultSATBaseDir = "/tmp/export/bee-sat"
 	t.Cleanup(func() { DefaultSATBaseDir = oldSATBaseDir })
--- a/audit/internal/app/support_bundle.go
+++ b/audit/internal/app/support_bundle.go
@@ -54,7 +54,7 @@ if ! command -v lspci >/dev/null 2>&1; then
  exit 0
 fi
 found=0
-for gpu in $(lspci -Dn | awk '$3 ~ /^10de:/ {print $1}'); do
+	for gpu in $(lspci -Dn | awk '$2 ~ /^03(00|02):$/ && $3 ~ /^10de:/ {print $1}'); do
  found=1
  echo "=== GPU $gpu ==="
  lspci -s "$gpu" -vv 2>&1 || true
@@ -73,8 +73,13 @@ fi
 	{name: "system/pcie-nvidia-link.txt", cmd: []string{"sh", "-c", `
 for d in /sys/bus/pci/devices/*/; do
  vendor=$(cat "$d/vendor" 2>/dev/null)
-  [ "$vendor" = "0x10de" ] || continue
-  dev=$(basename "$d")
+	  [ "$vendor" = "0x10de" ] || continue
+	  class=$(cat "$d/class" 2>/dev/null)
+	  case "$class" in
+	    0x030000|0x030200) ;;
+	    *) continue ;;
+	  esac
+	  dev=$(basename "$d")
  echo "=== $dev ==="
  for f in current_link_speed current_link_width max_link_speed max_link_width; do
    printf "  %-22s %s\n" "$f" "$(cat "$d/$f" 2>/dev/null)"
@@ -208,7 +213,7 @@ func BuildSupportBundle(exportDir string) (string, error) {

 	now := time.Now().UTC()
 	date := now.Format("2006-01-02")
-	tod := now.Format("15:04:05")
+	tod := now.Format("150405")
 	ver := bundleVersion()
 	model := serverModelForBundle()
 	sn := serverSerialForBundle()
--- a/audit/internal/collector/nic_mellanox.go
+++ b/audit/internal/collector/nic_mellanox.go
@@ -179,11 +179,3 @@ func commandOutputWithTimeout(timeout time.Duration, name string, args ...string
 	defer cancel()
 	return exec.CommandContext(ctx, name, args...).Output()
 }
-
-func interfaceHasCarrier(iface string) bool {
-	raw, err := readNetCarrierFile(iface)
-	if err != nil {
-		return false
-	}
-	return strings.TrimSpace(raw) == "1"
-}
--- a/audit/internal/collector/nic_telemetry.go
+++ b/audit/internal/collector/nic_telemetry.go
@@ -58,12 +58,10 @@ func enrichPCIeWithNICTelemetry(devs []schema.HardwarePCIeDevice) []schema.Hardw
 			}
 		}

-		if interfaceHasCarrier(iface) {
-			if out, err := ethtoolModuleQuery(iface); err == nil {
-				if injectSFPDOMTelemetry(&devs[i], out) {
-					enriched++
-					continue
-				}
+		if out, err := ethtoolModuleQuery(iface); err == nil {
+			if injectSFPDOMTelemetry(&devs[i], out) {
+				enriched++
+				continue
 			}
 		}
 		if len(devs[i].MacAddresses) > 0 || devs[i].Firmware != nil {
@@ -115,8 +113,38 @@ func injectSFPDOMTelemetry(dev *schema.HardwarePCIeDevice, raw string) bool {
 		}
 		key := strings.ToLower(strings.TrimSpace(trimmed[:idx]))
 		val := strings.TrimSpace(trimmed[idx+1:])
+		if val == "" || strings.EqualFold(val, "not supported") || strings.EqualFold(val, "unknown") {
+			continue
+		}

 		switch {
+		case key == "identifier":
+			s := parseSFPIdentifier(val)
+			dev.SFPIdentifier = &s
+			t := true
+			dev.SFPPresent = &t
+			changed = true
+		case key == "connector":
+			s := parseSFPConnector(val)
+			dev.SFPConnector = &s
+			changed = true
+		case key == "vendor name":
+			s := strings.TrimSpace(val)
+			dev.SFPVendor = &s
+			changed = true
+		case key == "vendor pn":
+			s := strings.TrimSpace(val)
+			dev.SFPPartNumber = &s
+			changed = true
+		case key == "vendor sn":
+			s := strings.TrimSpace(val)
+			dev.SFPSerialNumber = &s
+			changed = true
+		case strings.Contains(key, "laser wavelength"):
+			if f, ok := firstFloat(val); ok {
+				dev.SFPWavelengthNM = &f
+				changed = true
+			}
 		case strings.Contains(key, "module temperature"):
 			if f, ok := firstFloat(val); ok {
 				dev.SFPTemperatureC = &f
@@ -147,12 +175,61 @@ func injectSFPDOMTelemetry(dev *schema.HardwarePCIeDevice, raw string) bool {
 	return changed
 }

+// parseSFPIdentifier extracts the human-readable transceiver type from the
+// raw ethtool identifier line, e.g. "0x03 (SFP)" → "SFP".
+func parseSFPIdentifier(val string) string {
+	if s := extractParens(val); s != "" {
+		return s
+	}
+	return val
+}
+
+// parseSFPConnector extracts the connector type from the raw ethtool line,
+// e.g. "0x07 (LC)" → "LC".
+func parseSFPConnector(val string) string {
+	if s := extractParens(val); s != "" {
+		return s
+	}
+	return val
+}
+
+var parenRe = regexp.MustCompile(`\(([^)]+)\)`)
+
+func extractParens(s string) string {
+	m := parenRe.FindStringSubmatch(s)
+	if len(m) < 2 {
+		return ""
+	}
+	return strings.TrimSpace(m[1])
+}
+
 func parseSFPDOM(raw string) map[string]any {
 	dev := schema.HardwarePCIeDevice{}
 	if !injectSFPDOMTelemetry(&dev, raw) {
 		return map[string]any{}
 	}
 	out := map[string]any{}
+	if dev.SFPPresent != nil {
+		out["sfp_present"] = *dev.SFPPresent
+	}
+	if dev.SFPIdentifier != nil {
+		out["sfp_identifier"] = *dev.SFPIdentifier
+	}
+	if dev.SFPConnector != nil {
+		out["sfp_connector"] = *dev.SFPConnector
+	}
+	if dev.SFPVendor != nil {
+		out["sfp_vendor"] = *dev.SFPVendor
+	}
+	if dev.SFPPartNumber != nil {
+		out["sfp_part_number"] = *dev.SFPPartNumber
+	}
+	if dev.SFPSerialNumber != nil {
+		out["sfp_serial_number"] = *dev.SFPSerialNumber
+	}
+	if dev.SFPWavelengthNM != nil {
+		out["sfp_wavelength_nm"] = *dev.SFPWavelengthNM
+	}
 	if dev.SFPTemperatureC != nil {
 		out["sfp_temperature_c"] = *dev.SFPTemperatureC
 	}
--- a/audit/internal/collector/nic_telemetry_test.go
+++ b/audit/internal/collector/nic_telemetry_test.go
@@ -122,10 +122,7 @@ func TestEnrichPCIeWithNICTelemetrySkipsModuleQueryWithoutCarrier(t *testing.T)
 	readNetAddressFile = func(string) (string, error) { return "aa:bb:cc:dd:ee:ff", nil }
 	readNetCarrierFile = func(string) (string, error) { return "0", nil }
 	ethtoolInfoQuery = func(string) (string, error) { return "", fmt.Errorf("skip firmware") }
-	ethtoolModuleQuery = func(string) (string, error) {
-		t.Fatal("ethtool -m should not be called without carrier")
-		return "", nil
-	}
+	ethtoolModuleQuery = func(string) (string, error) { return "", fmt.Errorf("no module") }

 	class := "EthernetController"
 	bdf := "0000:18:00.0"
--- a/audit/internal/collector/nvidia.go
+++ b/audit/internal/collector/nvidia.go
@@ -15,6 +15,7 @@ const nvidiaVendorID = 0x10de
 type nvidiaGPUInfo struct {
 	Index              int
 	BDF                string
+	Name               string
 	Serial             string
 	VBIOS              string
 	TemperatureC       *float64
@@ -73,6 +74,9 @@ func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[str
 			continue
 		}

+		if v := strings.TrimSpace(info.Name); v != "" {
+			devs[i].Model = &v
+		}
 		if v := strings.TrimSpace(info.Serial); v != "" {
 			devs[i].SerialNumber = &v
 		}
@@ -99,7 +103,7 @@ func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[str
 func queryNVIDIAGPUs() (map[string]nvidiaGPUInfo, error) {
 	out, err := exec.Command(
 		"nvidia-smi",
-		"--query-gpu=index,pci.bus_id,serial,vbios_version,temperature.gpu,power.draw,ecc.errors.uncorrected.aggregate.total,ecc.errors.corrected.aggregate.total,clocks_throttle_reasons.hw_slowdown,pcie.link.gen.current,pcie.link.gen.max,pcie.link.width.current,pcie.link.width.max",
+		"--query-gpu=index,pci.bus_id,name,serial,vbios_version,temperature.gpu,power.draw,ecc.errors.uncorrected.aggregate.total,ecc.errors.corrected.aggregate.total,clocks_throttle_reasons.hw_slowdown,pcie.link.gen.current,pcie.link.gen.max,pcie.link.width.current,pcie.link.width.max",
 		"--format=csv,noheader,nounits",
 	).Output()
 	if err != nil {
@@ -123,8 +127,8 @@ func parseNVIDIASMIQuery(raw string) (map[string]nvidiaGPUInfo, error) {
 		if len(rec) == 0 {
 			continue
 		}
-		if len(rec) < 13 {
-			return nil, fmt.Errorf("unexpected nvidia-smi columns: got %d, want 13", len(rec))
+		if len(rec) < 14 {
+			return nil, fmt.Errorf("unexpected nvidia-smi columns: got %d, want 14", len(rec))
 		}

 		bdf := normalizePCIeBDF(rec[1])
@@ -135,17 +139,18 @@ func parseNVIDIASMIQuery(raw string) (map[string]nvidiaGPUInfo, error) {
 		info := nvidiaGPUInfo{
 			Index:              parseRequiredInt(rec[0]),
 			BDF:                bdf,
-			Serial:             strings.TrimSpace(rec[2]),
-			VBIOS:              strings.TrimSpace(rec[3]),
-			TemperatureC:       parseMaybeFloat(rec[4]),
-			PowerW:             parseMaybeFloat(rec[5]),
-			ECCUncorrected:     parseMaybeInt64(rec[6]),
-			ECCCorrected:       parseMaybeInt64(rec[7]),
-			HWSlowdown:         parseMaybeBool(rec[8]),
-			PCIeLinkGenCurrent: parseMaybeInt(rec[9]),
-			PCIeLinkGenMax:     parseMaybeInt(rec[10]),
-			PCIeLinkWidthCur:   parseMaybeInt(rec[11]),
-			PCIeLinkWidthMax:   parseMaybeInt(rec[12]),
+			Name:               strings.TrimSpace(rec[2]),
+			Serial:             strings.TrimSpace(rec[3]),
+			VBIOS:              strings.TrimSpace(rec[4]),
+			TemperatureC:       parseMaybeFloat(rec[5]),
+			PowerW:             parseMaybeFloat(rec[6]),
+			ECCUncorrected:     parseMaybeInt64(rec[7]),
+			ECCCorrected:       parseMaybeInt64(rec[8]),
+			HWSlowdown:         parseMaybeBool(rec[9]),
+			PCIeLinkGenCurrent: parseMaybeInt(rec[10]),
+			PCIeLinkGenMax:     parseMaybeInt(rec[11]),
+			PCIeLinkWidthCur:   parseMaybeInt(rec[12]),
+			PCIeLinkWidthMax:   parseMaybeInt(rec[13]),
 		}
 		result[bdf] = info
 	}
--- a/audit/internal/collector/nvidia_test.go
+++ b/audit/internal/collector/nvidia_test.go
@@ -6,7 +6,7 @@ import (
 )

 func TestParseNVIDIASMIQuery(t *testing.T) {
-	raw := "0, 00000000:65:00.0, GPU-SERIAL-1, 96.00.1F.00.02, 54, 210.33, 0, 5, Not Active, 4, 4, 16, 16\n"
+	raw := "0, 00000000:65:00.0, NVIDIA H100 80GB HBM3, GPU-SERIAL-1, 96.00.1F.00.02, 54, 210.33, 0, 5, Not Active, 4, 4, 16, 16\n"
 	byBDF, err := parseNVIDIASMIQuery(raw)
 	if err != nil {
 		t.Fatalf("parse failed: %v", err)
@@ -16,6 +16,9 @@ func TestParseNVIDIASMIQuery(t *testing.T) {
 	if !ok {
 		t.Fatalf("gpu by normalized bdf not found")
 	}
+	if gpu.Name != "NVIDIA H100 80GB HBM3" {
+		t.Fatalf("name: got %q", gpu.Name)
+	}
 	if gpu.Serial != "GPU-SERIAL-1" {
 		t.Fatalf("serial: got %q", gpu.Serial)
 	}
--- a/audit/internal/collector/pcie.go
+++ b/audit/internal/collector/pcie.go
@@ -2,6 +2,7 @@ package collector

 import (
 	"bee/audit/internal/schema"
+	"fmt"
 	"log/slog"
 	"os/exec"
 	"strconv"
@@ -79,6 +80,25 @@ func shouldIncludePCIeDevice(class, vendor, device string) bool {
 		}
 	}

+	// Exclude BMC/management virtual VGA adapters — these are firmware video chips,
+	// not real GPUs, and pollute the GPU inventory (e.g. iBMC, iDRAC, iLO VGA).
+	if strings.Contains(c, "vga") || strings.Contains(c, "display") || strings.Contains(c, "3d") {
+		bmcPatterns := []string{
+			"management system chip",
+			"management controller",
+			"ibmc",
+			"idrac",
+			"ilo vga",
+			"aspeed",
+			"matrox",
+		}
+		for _, bad := range bmcPatterns {
+			if strings.Contains(d, bad) {
+				return false
+			}
+		}
+	}
+
 	if strings.Contains(v, "advanced micro devices") || strings.Contains(v, "[amd]") {
 		internalAMDPatterns := []string{
 			"dummy function",
@@ -153,6 +173,9 @@ func parseLspciDevice(fields map[string]string) schema.HardwarePCIeDevice {

 	// SVendor/SDevice available but not in schema — skip

+	// Warn if PCIe link is running below its maximum negotiated speed.
+	applyPCIeLinkSpeedWarning(&dev)
+
 	return dev
 }

@@ -222,6 +245,41 @@ func readPCIStringAttribute(bdf, attribute string) (string, bool) {
 	return value, true
 }

+// applyPCIeLinkSpeedWarning sets the device status to Warning if the current PCIe link
+// speed is below the maximum negotiated speed supported by both ends.
+func applyPCIeLinkSpeedWarning(dev *schema.HardwarePCIeDevice) {
+	if dev.LinkSpeed == nil || dev.MaxLinkSpeed == nil {
+		return
+	}
+	if pcieLinkSpeedRank(*dev.LinkSpeed) < pcieLinkSpeedRank(*dev.MaxLinkSpeed) {
+		warn := statusWarning
+		dev.Status = &warn
+		desc := fmt.Sprintf("PCIe link speed degraded: running at %s, capable of %s", *dev.LinkSpeed, *dev.MaxLinkSpeed)
+		dev.ErrorDescription = &desc
+	}
+}
+
+// pcieLinkSpeedRank returns a numeric rank for a normalized Gen string (e.g. "Gen4" → 4).
+// Returns 0 for unrecognised values so comparisons fail safe.
+func pcieLinkSpeedRank(gen string) int {
+	switch gen {
+	case "Gen1":
+		return 1
+	case "Gen2":
+		return 2
+	case "Gen3":
+		return 3
+	case "Gen4":
+		return 4
+	case "Gen5":
+		return 5
+	case "Gen6":
+		return 6
+	default:
+		return 0
+	}
+}
+
 func normalizePCILinkSpeed(raw string) string {
 	raw = strings.TrimSpace(strings.ToLower(raw))
 	switch {
--- a/audit/internal/collector/pcie_filter_test.go
+++ b/audit/internal/collector/pcie_filter_test.go
@@ -1,6 +1,7 @@
 package collector

 import (
+	"bee/audit/internal/schema"
 	"encoding/json"
 	"strings"
 	"testing"
@@ -29,6 +30,8 @@ func TestShouldIncludePCIeDevice(t *testing.T) {
 		{name: "raid", class: "RAID bus controller", want: true},
 		{name: "nvme", class: "Non-Volatile memory controller", want: true},
 		{name: "vga", class: "VGA compatible controller", want: true},
+		{name: "ibmc vga", class: "VGA compatible controller", vendor: "Huawei Technologies Co., Ltd.", device: "Hi171x Series [iBMC Intelligent Management system chip w/VGA support]", want: false},
+		{name: "aspeed vga", class: "VGA compatible controller", vendor: "ASPEED Technology, Inc.", device: "ASPEED Graphics Family", want: false},
 		{name: "other encryption controller", class: "Encryption controller", vendor: "Intel Corporation", device: "QuickAssist", want: true},
 	}

@@ -139,3 +142,77 @@ func TestNormalizePCILinkSpeed(t *testing.T) {
 		}
 	}
 }
+
+func TestApplyPCIeLinkSpeedWarning(t *testing.T) {
+	ptr := func(s string) *string { return &s }
+
+	tests := []struct {
+		name        string
+		linkSpeed   *string
+		maxSpeed    *string
+		wantWarning bool
+		wantGenIn   string // substring expected in ErrorDescription when warning
+	}{
+		{
+			name:        "degraded Gen1 vs Gen5",
+			linkSpeed:   ptr("Gen1"),
+			maxSpeed:    ptr("Gen5"),
+			wantWarning: true,
+			wantGenIn:   "Gen1",
+		},
+		{
+			name:        "at max Gen5",
+			linkSpeed:   ptr("Gen5"),
+			maxSpeed:    ptr("Gen5"),
+			wantWarning: false,
+		},
+		{
+			name:        "degraded Gen4 vs Gen5",
+			linkSpeed:   ptr("Gen4"),
+			maxSpeed:    ptr("Gen5"),
+			wantWarning: true,
+			wantGenIn:   "Gen4",
+		},
+		{
+			name:        "missing current speed — no warning",
+			linkSpeed:   nil,
+			maxSpeed:    ptr("Gen5"),
+			wantWarning: false,
+		},
+		{
+			name:        "missing max speed — no warning",
+			linkSpeed:   ptr("Gen1"),
+			maxSpeed:    nil,
+			wantWarning: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			dev := schema.HardwarePCIeDevice{}
+			ok := statusOK
+			dev.Status = &ok
+			dev.LinkSpeed = tt.linkSpeed
+			dev.MaxLinkSpeed = tt.maxSpeed
+
+			applyPCIeLinkSpeedWarning(&dev)
+
+			gotWarn := dev.Status != nil && *dev.Status == statusWarning
+			if gotWarn != tt.wantWarning {
+				t.Fatalf("wantWarning=%v gotWarning=%v (status=%v)", tt.wantWarning, gotWarn, dev.Status)
+			}
+			if tt.wantWarning {
+				if dev.ErrorDescription == nil {
+					t.Fatal("expected ErrorDescription to be set")
+				}
+				if !strings.Contains(*dev.ErrorDescription, tt.wantGenIn) {
+					t.Fatalf("ErrorDescription %q does not contain %q", *dev.ErrorDescription, tt.wantGenIn)
+				}
+			} else {
+				if dev.ErrorDescription != nil {
+					t.Fatalf("unexpected ErrorDescription: %s", *dev.ErrorDescription)
+				}
+			}
+		})
+	}
+}
--- a/audit/internal/platform/benchmark.go
+++ b/audit/internal/platform/benchmark.go
@@ -335,11 +335,7 @@ func (s *System) RunNvidiaBenchmark(ctx context.Context, baseDir string, opts Nv
 		return "", fmt.Errorf("write summary.txt: %w", err)
 	}

-	archive := filepath.Join(baseDir, "gpu-benchmark-"+ts+".tar.gz")
-	if err := createTarGz(archive, runDir); err != nil {
-		return "", fmt.Errorf("pack benchmark archive: %w", err)
-	}
-	return archive, nil
+	return runDir, nil
 }

 func normalizeNvidiaBenchmarkOptionsForBenchmark(opts NvidiaBenchmarkOptions) NvidiaBenchmarkOptions {
--- a/audit/internal/platform/benchmark_report.go
+++ b/audit/internal/platform/benchmark_report.go
@@ -90,7 +90,7 @@ func renderBenchmarkReportWithCharts(result NvidiaBenchmarkResult, charts []benc
 	for _, gpu := range result.GPUs {
 		name := strings.TrimSpace(gpu.Name)
 		if name == "" {
-			name = "Unknown"
+			name = "Unknown GPU"
 		}
 		interconnect := "-"
 		if gpu.Scores.InterconnectScore > 0 {
--- a/audit/internal/platform/nvidia_stress.go
+++ b/audit/internal/platform/nvidia_stress.go
@@ -49,6 +49,9 @@ func buildNvidiaStressJob(opts NvidiaStressOptions) (satJob, error) {
 			"--seconds", strconv.Itoa(opts.DurationSec),
 			"--size-mb", strconv.Itoa(opts.SizeMB),
 		}
+		if opts.StaggerSeconds > 0 && len(selected) > 1 {
+			cmd = append(cmd, "--stagger-seconds", strconv.Itoa(opts.StaggerSeconds))
+		}
 		if len(selected) > 0 {
 			cmd = append(cmd, "--devices", joinIndexList(selected))
 		}
@@ -63,6 +66,9 @@ func buildNvidiaStressJob(opts NvidiaStressOptions) (satJob, error) {
 			"bee-john-gpu-stress",
 			"--seconds", strconv.Itoa(opts.DurationSec),
 		}
+		if opts.StaggerSeconds > 0 && len(selected) > 1 {
+			cmd = append(cmd, "--stagger-seconds", strconv.Itoa(opts.StaggerSeconds))
+		}
 		if len(selected) > 0 {
 			cmd = append(cmd, "--devices", joinIndexList(selected))
 		}
--- a/audit/internal/platform/platform_stress.go
+++ b/audit/internal/platform/platform_stress.go
@@ -161,13 +161,7 @@ func (s *System) RunPlatformStress(
 	}
 	_ = os.WriteFile(filepath.Join(runDir, "summary.txt"), []byte(summary), 0644)

-	// Pack tar.gz
-	archivePath := filepath.Join(baseDir, "platform-stress-"+stamp+".tar.gz")
-	if err := packPlatformDir(runDir, archivePath); err != nil {
-		return "", fmt.Errorf("pack archive: %w", err)
-	}
-	_ = os.RemoveAll(runDir)
-	return archivePath, nil
+	return runDir, nil
 }

 // collectPhase samples live metrics every second until ctx is done.
--- a/audit/internal/platform/runtime.go
+++ b/audit/internal/platform/runtime.go
@@ -1,6 +1,7 @@
 package platform

 import (
+	"bufio"
 	"os"
 	"os/exec"
 	"strings"
@@ -114,6 +115,8 @@ func (s *System) CollectRuntimeHealth(exportDir string) (schema.RuntimeHealth, e
 	}

 	s.collectGPURuntimeHealth(vendor, &health)
+	s.collectToRAMHealth(&health)
+	s.collectUSBExportHealth(&health)

 	if health.Status != "FAILED" && len(health.Issues) > 0 {
 		health.Status = "PARTIAL"
@@ -168,6 +171,90 @@ func resolvedToolStatus(display string, candidates ...string) ToolStatus {
 	return ToolStatus{Name: display}
 }

+// collectToRAMHealth checks whether the LiveCD ISO has been copied to RAM.
+// Status values: "ok" = in RAM, "warning" = toram not active (no copy attempted),
+// "failed" = toram was requested but medium is not in RAM (copy failed or in progress).
+func (s *System) collectToRAMHealth(health *schema.RuntimeHealth) {
+	inRAM := s.IsLiveMediaInRAM()
+	active := toramActive()
+	switch {
+	case inRAM:
+		health.ToRAMStatus = "ok"
+	case active:
+		// toram was requested but medium is not yet/no longer in RAM
+		health.ToRAMStatus = "failed"
+		health.Issues = append(health.Issues, schema.RuntimeIssue{
+			Code:        "toram_copy_failed",
+			Severity:    "warning",
+			Description: "toram boot parameter is set but the live medium is not mounted from RAM.",
+		})
+	default:
+		health.ToRAMStatus = "warning"
+	}
+}
+
+// collectUSBExportHealth scans /proc/mounts for a writable USB-backed filesystem
+// suitable for log export. Sets USBExportPath to the first match found.
+func (s *System) collectUSBExportHealth(health *schema.RuntimeHealth) {
+	health.USBExportPath = findUSBExportMount()
+}
+
+// findUSBExportMount returns the mount point of the first writable USB filesystem
+// found in /proc/mounts (vfat, exfat, ext2/3/4, ntfs) whose backing block device
+// has USB transport. Returns "" if none found.
+func findUSBExportMount() string {
+	f, err := os.Open("/proc/mounts")
+	if err != nil {
+		return ""
+	}
+	defer f.Close()
+
+	// fs types that are expected on USB export drives
+	exportFSTypes := map[string]bool{
+		"vfat":  true,
+		"exfat": true,
+		"ext2":  true,
+		"ext3":  true,
+		"ext4":  true,
+		"ntfs":  true,
+		"ntfs3": true,
+		"fuseblk": true,
+	}
+
+	scanner := bufio.NewScanner(f)
+	for scanner.Scan() {
+		// fields: device mountpoint fstype options dump pass
+		fields := strings.Fields(scanner.Text())
+		if len(fields) < 4 {
+			continue
+		}
+		device, mountPoint, fsType, options := fields[0], fields[1], fields[2], fields[3]
+		if !exportFSTypes[strings.ToLower(fsType)] {
+			continue
+		}
+		// Skip read-only mounts
+		opts := strings.Split(options, ",")
+		readOnly := false
+		for _, o := range opts {
+			if strings.TrimSpace(o) == "ro" {
+				readOnly = true
+				break
+			}
+		}
+		if readOnly {
+			continue
+		}
+		// Check USB transport via lsblk on the device
+		if !strings.HasPrefix(device, "/dev/") {
+			continue
+		}
+		if blockDeviceTransport(device) == "usb" {
+			return mountPoint
+		}
+	}
+	return ""
+}
+
 func (s *System) collectGPURuntimeHealth(vendor string, health *schema.RuntimeHealth) {
 	lsmodText := commandText("lsmod")

--- a/audit/internal/platform/sat.go
+++ b/audit/internal/platform/sat.go
@@ -384,25 +384,39 @@ func (s *System) RunNCCLTests(ctx context.Context, baseDir string, logFunc func(
 	), logFunc)
 }

-func (s *System) RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+func (s *System) RunNvidiaOfficialComputePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, staggerSec int, logFunc func(string)) (string, error) {
 	selected, err := resolveDCGMGPUIndices(gpuIndices)
 	if err != nil {
 		return "", err
 	}
-	profCmd, err := resolveDCGMProfTesterCommand("--no-dcgm-validation", "-t", "1004", "-d", strconv.Itoa(normalizeNvidiaBurnDuration(durationSec)))
-	if err != nil {
-		return "", err
+	var (
+		profCmd []string
+		profEnv []string
+	)
+	if staggerSec > 0 && len(selected) > 1 {
+		profCmd = []string{
+			"bee-dcgmproftester-staggered",
+			"--seconds", strconv.Itoa(normalizeNvidiaBurnDuration(durationSec)),
+			"--stagger-seconds", strconv.Itoa(staggerSec),
+			"--devices", joinIndexList(selected),
+		}
+	} else {
+		profCmd, err = resolveDCGMProfTesterCommand("--no-dcgm-validation", "-t", "1004", "-d", strconv.Itoa(normalizeNvidiaBurnDuration(durationSec)))
+		if err != nil {
+			return "", err
+		}
+		profEnv = nvidiaVisibleDevicesEnv(selected)
 	}
 	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-compute", withNvidiaPersistenceMode(
 		satJob{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
 		satJob{name: "02-dcgmi-version.log", cmd: []string{"dcgmi", "-v"}},
-		satJob{
-			name:       "03-dcgmproftester.log",
-			cmd:        profCmd,
-			env:        nvidiaVisibleDevicesEnv(selected),
-			collectGPU: true,
-			gpuIndices: selected,
-		},
+			satJob{
+				name:       "03-dcgmproftester.log",
+				cmd:        profCmd,
+				env:        profEnv,
+				collectGPU: true,
+				gpuIndices: selected,
+			},
 		satJob{name: "04-nvidia-smi-after.log", cmd: []string{"nvidia-smi", "--query-gpu=index,name,temperature.gpu,power.draw,utilization.gpu,memory.used,memory.total", "--format=csv,noheader,nounits"}},
 	), logFunc)
 }
@@ -648,11 +662,7 @@ func (s *System) RunStorageAcceptancePack(ctx context.Context, baseDir string, e
 	if err := os.WriteFile(filepath.Join(runDir, "summary.txt"), []byte(summary.String()), 0644); err != nil {
 		return "", err
 	}
-	archive := filepath.Join(baseDir, "storage-"+ts+".tar.gz")
-	if err := createTarGz(archive, runDir); err != nil {
-		return "", err
-	}
-	return archive, nil
+	return runDir, nil
 }

 type satJob struct {
@@ -838,11 +848,7 @@ func runAcceptancePackCtx(ctx context.Context, baseDir, prefix string, jobs []sa
 		}
 	}

-	archive := filepath.Join(baseDir, prefix+"-"+ts+".tar.gz")
-	if err := createTarGz(archive, runDir); err != nil {
-		return "", err
-	}
-	return archive, nil
+	return runDir, nil
 }

 func updateNvidiaGPUStatus(perGPU map[int]*nvidiaGPUStatusFile, idx int, status, jobName, detail string) {
@@ -905,7 +911,7 @@ func writeNvidiaGPUStatusFiles(runDir, overall string, perGPU map[int]*nvidiaGPU
 			entry.Health = "UNKNOWN"
 		}
 		if entry.Name == "" {
-			entry.Name = "unknown"
+			entry.Name = "Unknown GPU"
 		}
 		var body strings.Builder
 		fmt.Fprintf(&body, "gpu_index=%d\n", entry.Index)
--- a/audit/internal/platform/sat_fan_stress.go
+++ b/audit/internal/platform/sat_fan_stress.go
@@ -223,11 +223,7 @@ func (s *System) RunFanStressTest(ctx context.Context, baseDir string, opts FanS
 		return "", err
 	}

-	archive := filepath.Join(baseDir, "fan-stress-"+ts+".tar.gz")
-	if err := createTarGz(archive, runDir); err != nil {
-		return "", err
-	}
-	return archive, nil
+	return runDir, nil
 }

 func applyFanStressDefaults(opts *FanStressOptions) {
--- a/audit/internal/platform/techdump.go
+++ b/audit/internal/platform/techdump.go
@@ -20,6 +20,7 @@ var techDumpFixedCommands = []struct {
 	{Name: "dmidecode", Args: []string{"-t", "4"}, File: "dmidecode-type4.txt"},
 	{Name: "dmidecode", Args: []string{"-t", "17"}, File: "dmidecode-type17.txt"},
 	{Name: "lspci", Args: []string{"-vmm", "-D"}, File: "lspci-vmm.txt"},
+	{Name: "lspci", Args: []string{"-vvv"}, File: "lspci-vvv.txt"},
 	{Name: "lsblk", Args: []string{"-J", "-d", "-o", "NAME,TYPE,SIZE,SERIAL,MODEL,TRAN,HCTL"}, File: "lsblk.json"},
 	{Name: "sensors", Args: []string{"-j"}, File: "sensors.json"},
 	{Name: "ipmitool", Args: []string{"fru", "print"}, File: "ipmitool-fru.txt"},
--- a/audit/internal/platform/types.go
+++ b/audit/internal/platform/types.go
@@ -70,6 +70,7 @@ type NvidiaStressOptions struct {
 	Loader            string
 	GPUIndices        []int
 	ExcludeGPUIndices []int
+	StaggerSeconds    int
 }

 func New() *System {
--- a/audit/internal/schema/hardware.go
+++ b/audit/internal/schema/hardware.go
@@ -22,6 +22,10 @@ type RuntimeHealth struct {
 	CUDAReady     bool                   `json:"cuda_ready,omitempty"`
 	NvidiaGSPMode string                 `json:"nvidia_gsp_mode,omitempty"` // "gsp-on", "gsp-off", "gsp-stuck"
 	NetworkStatus string                 `json:"network_status,omitempty"`
+	// ToRAMStatus: "ok" (ISO in RAM), "warning" (toram not active), "failed" (toram active but copy failed)
+	ToRAMStatus   string `json:"toram_status,omitempty"`
+	// USBExportPath: mount point of the first writable USB drive found, empty if none.
+	USBExportPath string `json:"usb_export_path,omitempty"`
 	Issues        []RuntimeIssue         `json:"issues,omitempty"`
 	Tools         []RuntimeToolStatus    `json:"tools,omitempty"`
 	Services      []RuntimeServiceStatus `json:"services,omitempty"`
@@ -183,6 +187,13 @@ type HardwarePCIeDevice struct {
 	BatteryTemperatureC    *float64       `json:"battery_temperature_c,omitempty"`
 	BatteryVoltageV        *float64       `json:"battery_voltage_v,omitempty"`
 	BatteryReplaceRequired *bool          `json:"battery_replace_required,omitempty"`
+	SFPPresent             *bool          `json:"sfp_present,omitempty"`
+	SFPIdentifier          *string        `json:"sfp_identifier,omitempty"`
+	SFPConnector           *string        `json:"sfp_connector,omitempty"`
+	SFPVendor              *string        `json:"sfp_vendor,omitempty"`
+	SFPPartNumber          *string        `json:"sfp_part_number,omitempty"`
+	SFPSerialNumber        *string        `json:"sfp_serial_number,omitempty"`
+	SFPWavelengthNM        *float64       `json:"sfp_wavelength_nm,omitempty"`
 	SFPTemperatureC        *float64       `json:"sfp_temperature_c,omitempty"`
 	SFPTXPowerDBM          *float64       `json:"sfp_tx_power_dbm,omitempty"`
 	SFPRXPowerDBM          *float64       `json:"sfp_rx_power_dbm,omitempty"`
--- a/audit/internal/webui/api.go
+++ b/audit/internal/webui/api.go
@@ -482,12 +482,13 @@ func (h *handler) handleAPISATRun(target string) http.HandlerFunc {
 			return
 		}

-		var body struct {
-			Duration           int      `json:"duration"`
-			StressMode         bool     `json:"stress_mode"`
-			GPUIndices         []int    `json:"gpu_indices"`
-			ExcludeGPUIndices  []int    `json:"exclude_gpu_indices"`
-			Loader             string   `json:"loader"`
+			var body struct {
+				Duration           int      `json:"duration"`
+				StressMode         bool     `json:"stress_mode"`
+				GPUIndices         []int    `json:"gpu_indices"`
+				ExcludeGPUIndices  []int    `json:"exclude_gpu_indices"`
+				StaggerGPUStart    bool     `json:"stagger_gpu_start"`
+				Loader             string   `json:"loader"`
 			Profile            string   `json:"profile"`
 			DisplayName        string   `json:"display_name"`
 			PlatformComponents []string `json:"platform_components"`
@@ -503,12 +504,13 @@ func (h *handler) handleAPISATRun(target string) http.HandlerFunc {
 		if strings.TrimSpace(body.DisplayName) != "" {
 			name = body.DisplayName
 		}
-		params := taskParams{
-			Duration:           body.Duration,
-			StressMode:         body.StressMode,
-			GPUIndices:         body.GPUIndices,
-			ExcludeGPUIndices:  body.ExcludeGPUIndices,
-			Loader:             body.Loader,
+			params := taskParams{
+				Duration:           body.Duration,
+				StressMode:         body.StressMode,
+				GPUIndices:         body.GPUIndices,
+				ExcludeGPUIndices:  body.ExcludeGPUIndices,
+				StaggerGPUStart:    body.StaggerGPUStart,
+				Loader:             body.Loader,
 			BurnProfile:        body.Profile,
 			DisplayName:        body.DisplayName,
 			PlatformComponents: body.PlatformComponents,
--- a/audit/internal/webui/charts_svg.go
+++ b/audit/internal/webui/charts_svg.go
@@ -83,6 +83,10 @@ func renderMetricChartSVG(title string, labels []string, times []time.Time, data
 		}
 	}

+	// Downsample to at most ~1400 points (one per pixel) before building SVG.
+	times, datasets = downsampleTimeSeries(times, datasets, 1400)
+	pointCount = len(times)
+
 	statsLabel := chartStatsLabel(datasets)

 	legendItems := []metricChartSeries{}
@@ -196,6 +200,19 @@ func drawGPUOverviewChartSVG(title string, labels []string, times []time.Time, s
 		}
 	}

+	// Downsample to at most ~1400 points before building SVG.
+	{
+		datasets := make([][]float64, len(series))
+		for i := range series {
+			datasets[i] = series[i].Values
+		}
+		times, datasets = downsampleTimeSeries(times, datasets, 1400)
+		pointCount = len(times)
+		for i := range series {
+			series[i].Values = datasets[i]
+		}
+	}
+
 	scales := make([]chartScale, len(series))
 	for i := range series {
 		min, max := chartSeriesBounds(series[i].Values)
@@ -626,6 +643,87 @@ func writeTimelineBoundaries(b *strings.Builder, layout chartLayout, start, end
 	b.WriteString(`</g>` + "\n")
 }

+// downsampleTimeSeries reduces the time series to at most maxPts points using
+// min-max bucketing. Each bucket contributes the index of its min and max value
+// (using the first full-length dataset as the reference series). All parallel
+// datasets are sampled at those same indices so all series stay aligned.
+// If len(times) <= maxPts the inputs are returned unchanged.
+func downsampleTimeSeries(times []time.Time, datasets [][]float64, maxPts int) ([]time.Time, [][]float64) {
+	n := len(times)
+	if n <= maxPts || maxPts <= 0 {
+		return times, datasets
+	}
+	buckets := maxPts / 2
+	if buckets < 1 {
+		buckets = 1
+	}
+	// Use the first dataset that has the same length as times as the reference
+	// for deciding which two indices to keep per bucket.
+	var ref []float64
+	for _, ds := range datasets {
+		if len(ds) == n {
+			ref = ds
+			break
+		}
+	}
+	selected := make([]int, 0, maxPts)
+	bucketSize := float64(n) / float64(buckets)
+	for b := 0; b < buckets; b++ {
+		lo := int(math.Round(float64(b) * bucketSize))
+		hi := int(math.Round(float64(b+1) * bucketSize))
+		if hi > n {
+			hi = n
+		}
+		if lo >= hi {
+			continue
+		}
+		if ref == nil {
+			selected = append(selected, lo)
+			if hi-1 != lo {
+				selected = append(selected, hi-1)
+			}
+			continue
+		}
+		minIdx, maxIdx := lo, lo
+		for i := lo + 1; i < hi; i++ {
+			if ref[i] < ref[minIdx] {
+				minIdx = i
+			}
+			if ref[i] > ref[maxIdx] {
+				maxIdx = i
+			}
+		}
+		if minIdx <= maxIdx {
+			selected = append(selected, minIdx)
+			if maxIdx != minIdx {
+				selected = append(selected, maxIdx)
+			}
+		} else {
+			selected = append(selected, maxIdx)
+			if minIdx != maxIdx {
+				selected = append(selected, minIdx)
+			}
+		}
+	}
+	outTimes := make([]time.Time, len(selected))
+	for i, idx := range selected {
+		outTimes[i] = times[idx]
+	}
+	outDatasets := make([][]float64, len(datasets))
+	for d, ds := range datasets {
+		if len(ds) != n {
+			outDatasets[d] = ds
+			continue
+		}
+		out := make([]float64, len(selected))
+		for i, idx := range selected {
+			out[i] = ds[idx]
+		}
+		outDatasets[d] = out
+	}
+	return outTimes, outDatasets
+}
+
 func chartXForTime(ts, start, end time.Time, left, right int) float64 {
 	if !end.After(start) {
 		return float64(left+right) / 2
--- a/audit/internal/webui/pages.go
+++ b/audit/internal/webui/pages.go
@@ -317,106 +317,299 @@ func renderHardwareSummaryCard(opts HandlerOptions) string {
 	if err != nil {
 		return `<div class="card"><div class="card-head card-head-actions"><span>Hardware Summary</span><div class="card-head-buttons"><button class="btn btn-primary btn-sm" onclick="auditModalRun()">Run audit</button></div></div><div class="card-body"></div></div>`
 	}
-	// Parse just enough fields for the summary banner
-	var snap struct {
-		Summary struct {
-			CPU     struct{ Model string }
-			Memory  struct{ TotalGB float64 }
-			Storage []struct{ Device, Model, Size string }
-			GPUs    []struct{ Model string }
-			PSUs    []struct{ Model string }
-		}
-		Network struct {
-			Interfaces []struct {
-				Name  string
-				IPv4  []string
-				State string
-			}
-		}
-	}
-	// Try to extract top-level fields loosely
-	var raw map[string]json.RawMessage
-	if err := json.Unmarshal(data, &raw); err != nil {
+	var ingest schema.HardwareIngestRequest
+	if err := json.Unmarshal(data, &ingest); err != nil {
 		return `<div class="card"><div class="card-head">Hardware Summary</div><div class="card-body"><span class="badge badge-err">Parse error</span></div></div>`
 	}
-	_ = snap
+	hw := ingest.Hardware

-	// Also load runtime-health for badges
-	type componentHealth struct {
-		FailCount int `json:"fail_count"`
-		WarnCount int `json:"warn_count"`
+	var records []app.ComponentStatusRecord
+	if db, err := app.OpenComponentStatusDB(filepath.Join(opts.ExportDir, "component-status.json")); err == nil {
+		records = db.All()
 	}
-	type healthSummary struct {
-		CPU     componentHealth `json:"cpu"`
-		Memory  componentHealth `json:"memory"`
-		Storage componentHealth `json:"storage"`
-		GPU     componentHealth `json:"gpu"`
-		PSU     componentHealth `json:"psu"`
-		Network componentHealth `json:"network"`
-	}
-	var health struct {
-		HardwareHealth healthSummary `json:"hardware_health"`
-	}
-	if hdata, herr := loadSnapshot(filepath.Join(opts.ExportDir, "runtime-health.json")); herr == nil {
-		_ = json.Unmarshal(hdata, &health)
-	}
-
-	badge := func(h componentHealth) string {
-		if h.FailCount > 0 {
-			return `<span class="badge badge-err">FAIL</span>`
-		}
-		if h.WarnCount > 0 {
-			return `<span class="badge badge-warn">WARN</span>`
-		}
-		return `<span class="badge badge-ok">OK</span>`
-	}
-
-	// Extract readable strings from raw JSON
-	getString := func(key string) string {
-		v, ok := raw[key]
-		if !ok {
-			return ""
-		}
-		var s string
-		if err := json.Unmarshal(v, &s); err == nil {
-			return s
-		}
-		return ""
-	}
-
-	cpuModel := getString("cpu_model")
-	memStr := getString("memory_summary")
-	gpuSummary := getString("gpu_summary")

 	var b strings.Builder
 	b.WriteString(`<div class="card"><div class="card-head">Hardware Summary</div><div class="card-body">`)
 	b.WriteString(`<table style="width:auto">`)
 	writeRow := func(label, value, badgeHTML string) {
-		b.WriteString(fmt.Sprintf(`<tr><td style="padding:6px 14px 6px 0;font-weight:700;white-space:nowrap">%s</td><td style="padding:6px 0">%s</td><td style="padding:6px 0 6px 12px">%s</td></tr>`,
+		b.WriteString(fmt.Sprintf(`<tr><td style="padding:6px 14px 6px 0;font-weight:700;white-space:nowrap">%s</td><td style="padding:6px 0;color:var(--muted);font-size:13px">%s</td><td style="padding:6px 0 6px 12px">%s</td></tr>`,
 			html.EscapeString(label), html.EscapeString(value), badgeHTML))
 	}
-	if cpuModel != "" {
-		writeRow("CPU", cpuModel, badge(health.HardwareHealth.CPU))
-	} else {
-		writeRow("CPU", "—", badge(health.HardwareHealth.CPU))
+
+	cpuRow := aggregateComponentStatus("CPU", records, []string{"cpu:all"}, nil)
+	writeRow("CPU", hwDescribeCPU(hw), runtimeStatusBadge(cpuRow.Status))
+
+	memRow := aggregateComponentStatus("Memory", records, []string{"memory:all"}, []string{"memory:"})
+	writeRow("Memory", hwDescribeMemory(hw), runtimeStatusBadge(memRow.Status))
+
+	storageRow := aggregateComponentStatus("Storage", records, []string{"storage:all"}, []string{"storage:"})
+	writeRow("Storage", hwDescribeStorage(hw), runtimeStatusBadge(storageRow.Status))
+
+	gpuRow := aggregateComponentStatus("GPU", records, nil, []string{"pcie:gpu:"})
+	writeRow("GPU", hwDescribeGPU(hw), runtimeStatusBadge(gpuRow.Status))
+
+	psuRow := aggregateComponentStatus("PSU", records, nil, []string{"psu:"})
+	if psuRow.Status == "UNKNOWN" && len(hw.PowerSupplies) > 0 {
+		psuRow.Status = hwPSUStatus(hw.PowerSupplies)
 	}
-	if memStr != "" {
-		writeRow("Memory", memStr, badge(health.HardwareHealth.Memory))
-	} else {
-		writeRow("Memory", "—", badge(health.HardwareHealth.Memory))
+	writeRow("PSU", hwDescribePSU(hw), runtimeStatusBadge(psuRow.Status))
+
+	if nicDesc := hwDescribeNIC(hw); nicDesc != "" {
+		writeRow("Network", nicDesc, "")
 	}
-	if gpuSummary != "" {
-		writeRow("GPU", gpuSummary, badge(health.HardwareHealth.GPU))
-	} else {
-		writeRow("GPU", "—", badge(health.HardwareHealth.GPU))
-	}
-	writeRow("Storage", "—", badge(health.HardwareHealth.Storage))
-	writeRow("PSU", "—", badge(health.HardwareHealth.PSU))
+
 	b.WriteString(`</table>`)
 	b.WriteString(`</div></div>`)
 	return b.String()
 }

+// hwDescribeCPU returns a human-readable CPU summary, e.g. "2× Intel Xeon Gold 6338".
+func hwDescribeCPU(hw schema.HardwareSnapshot) string {
+	counts := map[string]int{}
+	order := []string{}
+	for _, cpu := range hw.CPUs {
+		model := "Unknown CPU"
+		if cpu.Model != nil && *cpu.Model != "" {
+			model = *cpu.Model
+		}
+		if counts[model] == 0 {
+			order = append(order, model)
+		}
+		counts[model]++
+	}
+	if len(order) == 0 {
+		return "—"
+	}
+	parts := make([]string, 0, len(order))
+	for _, m := range order {
+		if counts[m] > 1 {
+			parts = append(parts, fmt.Sprintf("%d× %s", counts[m], m))
+		} else {
+			parts = append(parts, m)
+		}
+	}
+	return strings.Join(parts, ", ")
+}
+
+// hwDescribeMemory returns a summary like "16× 32 GB DDR4".
+func hwDescribeMemory(hw schema.HardwareSnapshot) string {
+	type key struct {
+		sizeMB int
+		typ    string
+	}
+	counts := map[key]int{}
+	order := []key{}
+	for _, dimm := range hw.Memory {
+		if dimm.SizeMB == nil || *dimm.SizeMB == 0 {
+			continue
+		}
+		t := ""
+		if dimm.Type != nil {
+			t = *dimm.Type
+		}
+		k := key{*dimm.SizeMB, t}
+		if counts[k] == 0 {
+			order = append(order, k)
+		}
+		counts[k]++
+	}
+	if len(order) == 0 {
+		return "—"
+	}
+	parts := make([]string, 0, len(order))
+	for _, k := range order {
+		gb := k.sizeMB / 1024
+		desc := fmt.Sprintf("%d× %d GB", counts[k], gb)
+		if k.typ != "" {
+			desc += " " + k.typ
+		}
+		parts = append(parts, desc)
+	}
+	return strings.Join(parts, ", ")
+}
+
+// hwDescribeStorage returns a summary like "4× 3.84 TB NVMe, 2× 1.92 TB SATA".
+func hwDescribeStorage(hw schema.HardwareSnapshot) string {
+	type key struct {
+		sizeGB int
+		iface  string
+	}
+	counts := map[key]int{}
+	order := []key{}
+	for _, disk := range hw.Storage {
+		sz := 0
+		if disk.SizeGB != nil {
+			sz = *disk.SizeGB
+		}
+		iface := ""
+		if disk.Interface != nil {
+			iface = *disk.Interface
+		} else if disk.Type != nil {
+			iface = *disk.Type
+		}
+		k := key{sz, iface}
+		if counts[k] == 0 {
+			order = append(order, k)
+		}
+		counts[k]++
+	}
+	if len(order) == 0 {
+		return "—"
+	}
+	parts := make([]string, 0, len(order))
+	for _, k := range order {
+		var sizeStr string
+		if k.sizeGB >= 1000 {
+			sizeStr = fmt.Sprintf("%.2g TB", float64(k.sizeGB)/1000)
+		} else if k.sizeGB > 0 {
+			sizeStr = fmt.Sprintf("%d GB", k.sizeGB)
+		} else {
+			sizeStr = "?"
+		}
+		desc := fmt.Sprintf("%d× %s", counts[k], sizeStr)
+		if k.iface != "" {
+			desc += " " + k.iface
+		}
+		parts = append(parts, desc)
+	}
+	return strings.Join(parts, ", ")
+}
+
+// hwDescribeGPU returns a summary like "8× NVIDIA H100 80GB".
+func hwDescribeGPU(hw schema.HardwareSnapshot) string {
+	counts := map[string]int{}
+	order := []string{}
+	for _, dev := range hw.PCIeDevices {
+		if dev.DeviceClass == nil {
+			continue
+		}
+		if !isGPUDeviceClass(*dev.DeviceClass) {
+			continue
+		}
+		model := "Unknown GPU"
+		if dev.Model != nil && *dev.Model != "" {
+			model = *dev.Model
+		}
+		if counts[model] == 0 {
+			order = append(order, model)
+		}
+		counts[model]++
+	}
+	if len(order) == 0 {
+		return "—"
+	}
+	parts := make([]string, 0, len(order))
+	for _, m := range order {
+		if counts[m] > 1 {
+			parts = append(parts, fmt.Sprintf("%d× %s", counts[m], m))
+		} else {
+			parts = append(parts, m)
+		}
+	}
+	return strings.Join(parts, ", ")
+}
+
+// hwPSUStatus returns "OK", "CRITICAL", "WARNING", or "UNKNOWN" based on
+// PSU statuses from the audit snapshot. Used as fallback when component-status.json
+// has no psu: records yet (e.g. first boot before audit writes them).
+func hwPSUStatus(psus []schema.HardwarePowerSupply) string {
+	worst := "UNKNOWN"
+	for _, psu := range psus {
+		if psu.Status == nil {
+			continue
+		}
+		switch strings.ToUpper(strings.TrimSpace(*psu.Status)) {
+		case "CRITICAL":
+			return "CRITICAL"
+		case "WARNING":
+			if worst != "CRITICAL" {
+				worst = "WARNING"
+			}
+		case "OK":
+			if worst == "UNKNOWN" {
+				worst = "OK"
+			}
+		}
+	}
+	return worst
+}
+
+// hwDescribePSU returns a summary like "2× 1600 W" or "2× PSU".
+func hwDescribePSU(hw schema.HardwareSnapshot) string {
+	n := len(hw.PowerSupplies)
+	if n == 0 {
+		return "—"
+	}
+	// Try to get a consistent wattage
+	watt := 0
+	consistent := true
+	for _, psu := range hw.PowerSupplies {
+		if psu.WattageW == nil {
+			consistent = false
+			break
+		}
+		if watt == 0 {
+			watt = *psu.WattageW
+		} else if *psu.WattageW != watt {
+			consistent = false
+			break
+		}
+	}
+	if consistent && watt > 0 {
+		return fmt.Sprintf("%d× %d W", n, watt)
+	}
+	return fmt.Sprintf("%d× PSU", n)
+}
+
+// hwDescribeNIC returns a summary like "2× Mellanox ConnectX-6".
+func hwDescribeNIC(hw schema.HardwareSnapshot) string {
+	counts := map[string]int{}
+	order := []string{}
+	for _, dev := range hw.PCIeDevices {
+		isNIC := false
+		if dev.DeviceClass != nil {
+			c := strings.ToLower(strings.TrimSpace(*dev.DeviceClass))
+			isNIC = c == "ethernetcontroller" || c == "networkcontroller" || strings.Contains(c, "fibrechannel")
+		}
+		if !isNIC && len(dev.MacAddresses) == 0 {
+			continue
+		}
+		model := ""
+		if dev.Model != nil && *dev.Model != "" {
+			model = *dev.Model
+		} else if dev.Manufacturer != nil && *dev.Manufacturer != "" {
+			model = *dev.Manufacturer + " NIC"
+		} else {
+			model = "NIC"
+		}
+		if counts[model] == 0 {
+			order = append(order, model)
+		}
+		counts[model]++
+	}
+	if len(order) == 0 {
+		return ""
+	}
+	parts := make([]string, 0, len(order))
+	for _, m := range order {
+		if counts[m] > 1 {
+			parts = append(parts, fmt.Sprintf("%d× %s", counts[m], m))
+		} else {
+			parts = append(parts, m)
+		}
+	}
+	return strings.Join(parts, ", ")
+}
+
+func isGPUDeviceClass(class string) bool {
+	switch strings.TrimSpace(class) {
+	case "VideoController", "DisplayController", "ProcessingAccelerator":
+		return true
+	default:
+		return false
+	}
+}
+
 func renderAuditModal() string {
 	return `<div id="audit-modal-overlay" style="display:none;position:fixed;inset:0;background:rgba(0,0,0,.5);z-index:100;align-items:center;justify-content:center">
  <div style="background:#fff;border-radius:6px;padding:24px;min-width:480px;max-width:1100px;width:min(1100px,92vw);max-height:92vh;overflow:auto;position:relative">
@@ -481,8 +674,9 @@ func renderHealthCard(opts HandlerOptions) string {
 		buildRuntimeAccelerationRow(health),
 		buildRuntimeToolsRow(health),
 		buildRuntimeServicesRow(health),
+		buildRuntimeUSBExportRow(health),
+		buildRuntimeToRAMRow(health),
 	}
-	rows = append(rows, buildHardwareComponentRows(opts.ExportDir)...)
 	b.WriteString(`<table><thead><tr><th>Check</th><th>Status</th><th>Source</th><th>Issue</th></tr></thead><tbody>`)
 	for _, row := range rows {
 		b.WriteString(`<tr><td>` + html.EscapeString(row.Title) + `</td><td>` + runtimeStatusBadge(row.Status) + `</td><td>` + html.EscapeString(row.Source) + `</td><td>` + rowIssueHTML(row.Issue) + `</td></tr>`)
@@ -578,7 +772,13 @@ func buildRuntimeServicesRow(health schema.RuntimeHealth) runtimeHealthRow {
 	nonActive := make([]string, 0)
 	for _, svc := range health.Services {
 		state := strings.TrimSpace(strings.ToLower(svc.Status))
-		if state != "active" {
+		// "activating" and "deactivating" are transient states for oneshot services
+		// (RemainAfterExit=yes) — the service is running normally, not failed.
+		// Only "failed" and "inactive" (after services should be running) are problems.
+		switch state {
+		case "active", "activating", "deactivating", "reloading":
+			// OK — service is running or transitioning normally
+		default:
 			nonActive = append(nonActive, svc.Name+"="+svc.Status)
 		}
 	}
@@ -591,6 +791,51 @@ func buildRuntimeServicesRow(health schema.RuntimeHealth) runtimeHealthRow {
 	return runtimeHealthRow{Title: "Bee Services", Status: status, Source: "ServiceState", Issue: issue}
 }

+func buildRuntimeUSBExportRow(health schema.RuntimeHealth) runtimeHealthRow {
+	path := strings.TrimSpace(health.USBExportPath)
+	if path != "" {
+		return runtimeHealthRow{
+			Title:  "USB Export Drive",
+			Status: "OK",
+			Source: "/proc/mounts + lsblk",
+			Issue:  path,
+		}
+	}
+	return runtimeHealthRow{
+		Title:  "USB Export Drive",
+		Status: "WARNING",
+		Source: "/proc/mounts + lsblk",
+		Issue:  "No writable USB drive mounted. Plug in a USB drive to enable log export.",
+	}
+}
+
+func buildRuntimeToRAMRow(health schema.RuntimeHealth) runtimeHealthRow {
+	switch strings.ToLower(strings.TrimSpace(health.ToRAMStatus)) {
+	case "ok":
+		return runtimeHealthRow{
+			Title:  "LiveCD in RAM",
+			Status: "OK",
+			Source: "live-boot / /proc/mounts",
+			Issue:  "",
+		}
+	case "failed":
+		return runtimeHealthRow{
+			Title:  "LiveCD in RAM",
+			Status: "FAILED",
+			Source: "live-boot / /proc/mounts",
+			Issue:  "toram boot parameter set but ISO is not mounted from RAM. Copy may have failed.",
+		}
+	default:
+		// toram not active — ISO still on original boot media (USB/CD)
+		return runtimeHealthRow{
+			Title:  "LiveCD in RAM",
+			Status: "WARNING",
+			Source: "live-boot / /proc/mounts",
+			Issue:  "ISO not copied to RAM. Use \u201cCopy to RAM\u201d to free the boot drive and improve performance.",
+		}
+	}
+}
+
 func buildHardwareComponentRows(exportDir string) []runtimeHealthRow {
 	path := filepath.Join(exportDir, "component-status.json")
 	db, err := app.OpenComponentStatusDB(path)
@@ -1031,25 +1276,26 @@ func renderValidate(opts HandlerOptions) string {
 	return `<div class="alert alert-info" style="margin-bottom:16px"><strong>Non-destructive:</strong> Validate tests collect diagnostics only. They do not write to disks, do not run sustained load, and do not increment hardware wear counters.</div>
 <p style="color:var(--muted);font-size:13px;margin-bottom:16px">Tasks continue in the background — view progress in <a href="/tasks">Tasks</a>.</p>

-<div class="card" style="margin-bottom:16px">
-  <div class="card-head">Validate Profile</div>
-  <div class="card-body validate-profile-body">
-    <div class="validate-profile-col">
-      <div class="form-row" style="margin:0"><label>Cycles</label><input type="number" id="sat-cycles" value="1" min="1" max="100" style="width:100%"></div>
-      <div class="form-row" style="margin:12px 0 0"><label>Mode</label></div>
-      <label class="cb-row"><input type="radio" name="sat-mode" id="sat-mode-validate" value="validate" checked onchange="satModeChanged()"><span>Validate — quick non-destructive check</span></label>
-      <label class="cb-row"><input type="radio" name="sat-mode" id="sat-mode-stress" value="stress" onchange="satModeChanged()"><span>Stress — thorough load test (~30–60 min)</span></label>
-    </div>
-    <div class="validate-profile-col validate-profile-action">
-      <p style="color:var(--muted);font-size:12px;margin:0 0 10px">Runs validate modules sequentially with the selected cycle count and mode. Validate is quick (~5–15 min total); Stress is thorough (~30–60 min total).</p>
-      <button class="btn btn-primary" onclick="runAllSAT()">Validate one by one</button>
-    </div>
-    <div class="validate-profile-col"></div>
-  </div>
-  <div class="card-body" style="padding-top:0;display:flex;justify-content:center">
-    <span id="sat-all-status" style="font-size:12px;color:var(--muted)"></span>
-  </div>
-</div>
+	<div class="card" style="margin-bottom:16px">
+	  <div class="card-head">Validate Profile</div>
+	  <div class="card-body validate-profile-body">
+	    <div class="validate-profile-col">
+	      <div class="form-row" style="margin:0"><label>Cycles</label><input type="number" id="sat-cycles" value="1" min="1" max="100" style="width:100%"></div>
+	    </div>
+	    <div class="validate-profile-col">
+	      <div class="form-row" style="margin:12px 0 0"><label>Mode</label></div>
+	      <label class="cb-row"><input type="radio" name="sat-mode" id="sat-mode-validate" value="validate" checked onchange="satModeChanged()"><span>Validate — quick non-destructive check</span></label>
+	      <label class="cb-row"><input type="radio" name="sat-mode" id="sat-mode-stress" value="stress" onchange="satModeChanged()"><span>Stress — thorough load test (~30–60 min)</span></label>
+	    </div>
+	    <div class="validate-profile-col validate-profile-action">
+	      <p style="color:var(--muted);font-size:12px;margin:0 0 10px">Runs validate modules sequentially with the selected cycle count and mode. Validate is quick (~5–15 min total); Stress is thorough (~30–60 min total).</p>
+	      <button type="button" class="btn btn-primary" onclick="runAllSAT()">Validate one by one</button>
+	      <div style="margin-top:12px">
+	        <span id="sat-all-status" style="font-size:12px;color:var(--muted)"></span>
+	      </div>
+	    </div>
+	  </div>
+	</div>

 <div class="grid3">
 ` + renderSATCard("cpu", "CPU", "runSAT('cpu')", "", renderValidateCardBody(
@@ -1156,7 +1402,7 @@ func renderValidate(opts HandlerOptions) string {
 </div>
 <style>
 .validate-profile-body { display:grid; grid-template-columns:1fr 1fr 1fr; gap:24px; align-items:stretch; }
-.validate-profile-col { min-width:0; }
+.validate-profile-col { min-width:0; display:flex; flex-direction:column; }
 .validate-profile-action { display:flex; flex-direction:column; align-items:center; justify-content:center; }
 .validate-card-body { padding:0; }
 .validate-card-section { padding:12px 16px 0; }
@@ -1437,8 +1683,8 @@ function runAllSAT() {
  const cycles = Math.max(1, parseInt(document.getElementById('sat-cycles').value)||1);
  const status = document.getElementById('sat-all-status');
  status.textContent = 'Enqueuing...';
-  const stressOnlyTargets = ['nvidia-targeted-stress', 'nvidia-targeted-power', 'nvidia-pulse', 'nvidia-interconnect', 'nvidia-bandwidth', 'hpl'];
-  const baseTargets = ['nvidia','nvidia-targeted-stress','nvidia-targeted-power','nvidia-pulse','nvidia-interconnect','nvidia-bandwidth','hpl','memory','storage','cpu'].concat(selectedAMDValidateTargets());
+  const stressOnlyTargets = ['nvidia-targeted-stress', 'nvidia-targeted-power', 'nvidia-pulse', 'nvidia-interconnect', 'nvidia-bandwidth'];
+  const baseTargets = ['nvidia','nvidia-targeted-stress','nvidia-targeted-power','nvidia-pulse','nvidia-interconnect','nvidia-bandwidth','memory','storage','cpu'].concat(selectedAMDValidateTargets());
  const activeTargets = baseTargets.filter(target => {
    if (stressOnlyTargets.indexOf(target) >= 0 && !satStressMode()) return false;
    const btn = document.getElementById('sat-btn-' + target);
@@ -1612,6 +1858,11 @@ func formatValidateDeviceSummary(total int, models map[string]int, unit string)
 	if total != 1 {
 		label += "s"
 	}
+	// If there is only one model the leading count duplicates the per-model
+	// count already in parts (e.g. "4 GPU: 4 x RTX …" → "4 x RTX …").
+	if len(parts) == 1 {
+		return parts[0] + " " + label
+	}
 	return fmt.Sprintf("%d %s: %s", total, label, strings.Join(parts, ", "))
 }

@@ -2095,11 +2346,11 @@ func renderBurn() string {
      <label class="cb-row"><input type="radio" name="burn-profile" value="overnight"><span>Overnight — 8 hours</span></label>
    </div>
    <div class="burn-profile-col burn-profile-action">
-      <button class="btn btn-primary" onclick="runAllBurnTasks()">Burn one by one</button>
+      <button type="button" class="btn btn-primary" onclick="runAllBurnTasks()">Burn one by one</button>
      <p>Run checked tests one by one. Tests run without cooldown. Each test duration is determined by the Burn Profile. Total test duration is the sum of all selected tests multiplied by the Burn Profile duration.</p>
    </div>
    <div class="burn-profile-col burn-profile-action">
-      <button class="btn btn-secondary" onclick="runPlatformStress()">Thermal Cycling</button>
+      <button type="button" class="btn btn-secondary" onclick="runPlatformStress()">Thermal Cycling</button>
      <p>Run checked core test modules (CPU, MEM, GPU). Tests start at the same time and run for a period with short cooldown phases to stress the server cooling system.</p>
    </div>
  </div>
@@ -2116,12 +2367,16 @@ func renderBurn() string {
      <button class="btn btn-sm btn-secondary" type="button" onclick="burnSelectAll()">Select All</button>
      <button class="btn btn-sm btn-secondary" type="button" onclick="burnSelectNone()">Clear</button>
    </div>
-    <div id="burn-gpu-list" style="border:1px solid var(--border);border-radius:4px;padding:12px;min-height:88px">
-      <p style="color:var(--muted);font-size:13px">Loading NVIDIA GPUs...</p>
-    </div>
-    <p id="burn-selection-note" style="font-size:12px;color:var(--muted);margin:10px 0 0">Select at least one NVIDIA GPU to enable NVIDIA burn recipes.</p>
-  </div>
-</div>
+	    <div id="burn-gpu-list" style="border:1px solid var(--border);border-radius:4px;padding:12px;min-height:88px">
+	      <p style="color:var(--muted);font-size:13px">Loading NVIDIA GPUs...</p>
+	    </div>
+	    <p id="burn-selection-note" style="font-size:12px;color:var(--muted);margin:10px 0 0">Select at least one NVIDIA GPU to enable NVIDIA burn recipes.</p>
+	    <label class="cb-row" style="margin-top:10px">
+	      <input type="checkbox" id="burn-stagger-nvidia">
+	      <span>Ramp selected NVIDIA GPUs one by one before full-load hold. Uses a 3-minute stabilization window per GPU, then keeps all selected GPUs under load for the chosen Burn Profile duration.</span>
+	    </label>
+	  </div>
+	</div>

 <div class="burn-section">Core Burn Paths</div>
 <div class="grid2 burn-grid" style="margin-bottom:16px">
@@ -2147,10 +2402,6 @@ func renderBurn() string {
 </div>
 </div>

-<div class="burn-section">GPU-Specific Tests</div>
-<div class="grid2 burn-grid" style="margin-bottom:16px">
-</div>
-
 <div id="bi-output" style="display:none;margin-top:16px" class="card">
  <div class="card-head">Output <span id="bi-title"></span></div>
  <div class="card-body"><div id="bi-terminal" class="terminal"></div></div>
@@ -2199,6 +2450,11 @@ function burnSelectedGPUIndices() {
    .sort(function(a, b) { return a - b; });
 }

+function burnUseNvidiaRampUp() {
+  const el = document.getElementById('burn-stagger-nvidia');
+  return !!(el && el.checked);
+}
+
 function burnUpdateSelectionNote() {
  const note = document.getElementById('burn-selection-note');
  const selected = burnSelectedGPUIndices();
@@ -2258,6 +2514,9 @@ function enqueueBurnTask(target, label, extra, useSelectedNvidia) {
      return Promise.reject(new Error('Select at least one NVIDIA GPU.'));
    }
    body.gpu_indices = selected;
+    if (burnUseNvidiaRampUp() && selected.length > 1) {
+      body.stagger_gpu_start = true;
+    }
  }
  return fetch('/api/sat/' + target + '/run', {
    method: 'POST',
--- a/audit/internal/webui/server_test.go
+++ b/audit/internal/webui/server_test.go
@@ -1094,6 +1094,7 @@ func TestDashboardRendersRuntimeHealthTable(t *testing.T) {
 	}
 	body := rec.Body.String()
 	for _, needle := range []string{
+		// Runtime Health card — LiveCD checks only
 		`Runtime Health`,
 		`<th>Check</th><th>Status</th><th>Source</th><th>Issue</th>`,
 		`Export Directory`,
@@ -1102,16 +1103,18 @@ func TestDashboardRendersRuntimeHealthTable(t *testing.T) {
 		`CUDA / ROCm`,
 		`Required Utilities`,
 		`Bee Services`,
-		`<td>CPU</td>`,
-		`<td>Memory</td>`,
-		`<td>Storage</td>`,
-		`<td>GPU</td>`,
 		`CUDA runtime is not ready for GPU SAT.`,
 		`Missing: nvidia-smi`,
 		`bee-nvidia=inactive`,
-		`cpu SAT: FAILED`,
-		`storage SAT: FAILED`,
-		`sat:nvidia`,
+		// Hardware Summary card — component health badges
+		`Hardware Summary`,
+		`>CPU<`,
+		`>Memory<`,
+		`>Storage<`,
+		`>GPU<`,
+		`>PSU<`,
+		`badge-warn`,   // cpu Warning badge
+		`badge-err`,    // storage Critical badge
 	} {
 		if !strings.Contains(body, needle) {
 			t.Fatalf("dashboard missing %q: %s", needle, body)
--- a/audit/internal/webui/tasks.go
+++ b/audit/internal/webui/tasks.go
@@ -118,6 +118,7 @@ type taskParams struct {
 	StressMode         bool     `json:"stress_mode,omitempty"`
 	GPUIndices         []int    `json:"gpu_indices,omitempty"`
 	ExcludeGPUIndices  []int    `json:"exclude_gpu_indices,omitempty"`
+	StaggerGPUStart    bool     `json:"stagger_gpu_start,omitempty"`
 	SizeMB             int      `json:"size_mb,omitempty"`
 	Passes             int      `json:"passes,omitempty"`
 	Loader             string   `json:"loader,omitempty"`
@@ -162,6 +163,13 @@ func resolveBurnPreset(profile string) burnPreset {
 	}
 }

+func boolToNvidiaStaggerSeconds(enabled bool, selected []int) int {
+	if enabled && len(selected) > 1 {
+		return 180
+	}
+	return 0
+}
+
 func resolvePlatformStressPreset(profile string) platform.PlatformStressOptions {
 	acceptanceCycles := []platform.PlatformStressCycle{
 		{LoadSec: 85, IdleSec: 5},
@@ -592,7 +600,7 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 			RunNCCL:           t.params.RunNCCL,
 			ParallelGPUs:      t.params.ParallelGPUs,
 		}, j.append)
-	case "nvidia-compute":
+		case "nvidia-compute":
 		if a == nil {
 			err = fmt.Errorf("app not configured")
 			break
@@ -601,7 +609,11 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 		if t.params.BurnProfile != "" && dur <= 0 {
 			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
 		}
-		archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, j.append)
+			staggerSec := boolToNvidiaStaggerSeconds(t.params.StaggerGPUStart, t.params.GPUIndices)
+			if staggerSec > 0 {
+				j.append(fmt.Sprintf("NVIDIA staggered ramp-up enabled: %ds per GPU", staggerSec))
+			}
+			archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, staggerSec, j.append)
 	case "nvidia-targeted-power":
 		if a == nil {
 			err = fmt.Errorf("app not configured")
@@ -651,12 +663,13 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 		if t.params.BurnProfile != "" && dur <= 0 {
 			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
 		}
-		archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
-			DurationSec:       dur,
-			Loader:            t.params.Loader,
-			GPUIndices:        t.params.GPUIndices,
-			ExcludeGPUIndices: t.params.ExcludeGPUIndices,
-		}, j.append)
+			archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
+				DurationSec:       dur,
+				Loader:            t.params.Loader,
+				GPUIndices:        t.params.GPUIndices,
+				ExcludeGPUIndices: t.params.ExcludeGPUIndices,
+				StaggerSeconds:    boolToNvidiaStaggerSeconds(t.params.StaggerGPUStart, t.params.GPUIndices),
+			}, j.append)
 	case "memory":
 		if a == nil {
 			err = fmt.Errorf("app not configured")
--- a/bible-local/docs/gpu-model-propagation.md
+++ b/bible-local/docs/gpu-model-propagation.md
@@ -0,0 +1,117 @@
+# GPU Model Name Propagation
+
+How GPU model names are detected, stored, and displayed throughout the project.
+
+---
+
+## Detection Sources
+
+There are **two separate pipelines** for GPU model names — they use different structs and don't share state.
+
+### Pipeline A — Live / SAT (nvidia-smi query at runtime)
+
+**File:** `audit/internal/platform/sat.go`
+
+- `ListNvidiaGPUs()` → `NvidiaGPU.Name` (field: `name`, from `nvidia-smi --query-gpu=index,name,...`)
+- `ListNvidiaGPUStatuses()` → `NvidiaGPUStatus.Name`
+- Used by: GPU selection UI, live metrics labels, burn/stress test logic
+
+### Pipeline B — Benchmark results
+
+**File:** `audit/internal/platform/benchmark.go`, line 124
+
+- `queryBenchmarkGPUInfo(selected)` → `benchmarkGPUInfo.Name`
+- Stored in `BenchmarkGPUResult.Name` (`json:"name,omitempty"`)
+- Used by: benchmark history table, benchmark report
+
+### Pipeline C — Hardware audit JSON (PCIe schema)
+
+**File:** `audit/internal/schema/hardware.go`
+
+- `HardwarePCIeDevice.Model *string` (field name is **Model**, not Name)
+- For AMD GPUs: populated by `audit/internal/collector/amdgpu.go` from `info.Product`
+- For NVIDIA GPUs: **NOT populated** by `audit/internal/collector/nvidia.go` — the NVIDIA enricher sets telemetry/status but skips the Model field
+- Used by: hardware summary page (`hwDescribeGPU` in `pages.go:487`)
+
+---
+
+## Key Inconsistency: NVIDIA PCIe Model is Never Set
+
+`audit/internal/collector/nvidia.go` — `enrichPCIeWithNVIDIAData()` enriches NVIDIA PCIe devices with telemetry and status but does **not** populate `HardwarePCIeDevice.Model`.
+
+This means:
+- Hardware summary page shows "Unknown GPU" for all NVIDIA devices (falls back at `pages.go:486`)
+- AMD GPUs do have their model populated
+
+The fix would be: copy `gpu.Name` from the SAT pipeline into `dev.Model` inside `enrichPCIeWithNVIDIAData`.
+
+---
+
+## Benchmark History "Unknown GPU" Issue
+
+**Symptom:** Benchmark history table shows "GPU #N — Unknown GPU" columns instead of real GPU model names.
+
+**Root cause:** `BenchmarkGPUResult.Name` has tag `json:"name,omitempty"`. If `queryBenchmarkGPUInfo()` fails (warns at `benchmark.go:126`) or returns empty names, the Name field is never set and is omitted from JSON. Loaded results have empty Name → falls back to "Unknown GPU" at `pages.go:2226, 2237`.
+
+This happens for:
+- Older result files saved before the `Name` field was added
+- Runs where nvidia-smi query failed before the benchmark started
+
+---
+
+## Fallback Strings — Current State
+
+| Location | File | Fallback string |
+|---|---|---|
+| Hardware summary (PCIe) | `pages.go:486` | `"Unknown GPU"` |
+| Benchmark report summary | `benchmark_report.go:43` | `"Unknown GPU"` |
+| Benchmark report scorecard | `benchmark_report.go:93` | `"Unknown"` ← inconsistent |
+| Benchmark report detail | `benchmark_report.go:122` | `"Unknown GPU"` |
+| Benchmark history per-GPU col | `pages.go:2226` | `"Unknown GPU"` |
+| Benchmark history parallel col | `pages.go:2237` | `"Unknown GPU"` |
+| SAT status file write | `sat.go:922` | `"unknown"` ← lowercase, inconsistent |
+| GPU selection API | `api.go:163` | `"GPU N"` (no "Unknown") |
+
+**Rule:** all UI fallbacks should use `"Unknown GPU"`. The two outliers are `benchmark_report.go:93` (`"Unknown"`) and `sat.go:922` (`"unknown"`).
+
+---
+
+## GPU Selection UI
+
+**File:** `audit/internal/webui/pages.go`
+
+- Source: `GET /api/gpus` → `api.go` → `ListNvidiaGPUs()` → live nvidia-smi
+- Render: `'GPU ' + gpu.index + ' — ' + gpu.name + ' · ' + mem`
+- Fallback: `gpu.name || 'GPU ' + idx` (JS, line ~1432)
+
+This always shows the correct model because it queries nvidia-smi live. It is **not** connected to benchmark result data.
+
+---
+
+## Data Flow Summary
+
+```
+nvidia-smi (live)
+  └─ ListNvidiaGPUs() → NvidiaGPU.Name
+       ├─ GPU selection UI (always correct)
+       ├─ Live metrics labels (charts_svg.go)
+       └─ SAT/burn status file (sat.go)
+
+nvidia-smi (at benchmark start)
+  └─ queryBenchmarkGPUInfo() → benchmarkGPUInfo.Name
+       └─ BenchmarkGPUResult.Name (json:"name,omitempty")
+            ├─ Benchmark report
+            └─ Benchmark history table columns
+
+nvidia-smi / lspci (audit collection)
+  └─ HardwarePCIeDevice.Model (NVIDIA: NOT populated; AMD: populated)
+       └─ Hardware summary page hwDescribeGPU()
+```
+
+---
+
+## What Needs Fixing
+
+1. **NVIDIA PCIe Model** — `enrichPCIeWithNVIDIAData()` should set `dev.Model = &gpu.Name`
+2. **Fallback consistency** — `benchmark_report.go:93` should say `"Unknown GPU"` not `"Unknown"`; `sat.go:922` should say `"Unknown GPU"` not `"unknown"`
+3. **Old benchmark JSONs** — no fix possible for already-saved results with missing names (display-only issue)
--- a/iso/builder/config/bootloaders/grub-pc/grub.cfg
+++ b/iso/builder/config/bootloaders/grub-pc/grub.cfg
@@ -11,18 +11,18 @@ echo "  Hardware Audit LiveCD"
 echo ""

 menuentry "EASY-BEE" {
-    linux   @KERNEL_LIVE@ @APPEND_LIVE@ nomodeset bee.nvidia.mode=normal net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable nowatchdog nosoftlockup
+    linux   @KERNEL_LIVE@ @APPEND_LIVE@ nomodeset bee.nvidia.mode=normal net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1 nowatchdog nosoftlockup
    initrd  @INITRD_LIVE@
 }

 submenu "EASY-BEE (advanced options) -->" {
    menuentry "EASY-BEE — GSP=off" {
-        linux   @KERNEL_LIVE@ @APPEND_LIVE@ nomodeset bee.nvidia.mode=gsp-off net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable nowatchdog nosoftlockup
+        linux   @KERNEL_LIVE@ @APPEND_LIVE@ nomodeset bee.nvidia.mode=gsp-off net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1 nowatchdog nosoftlockup
        initrd  @INITRD_LIVE@
    }

    menuentry "EASY-BEE — KMS (no nomodeset)" {
-        linux   @KERNEL_LIVE@ @APPEND_LIVE@ bee.nvidia.mode=normal net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable nowatchdog nosoftlockup
+        linux   @KERNEL_LIVE@ @APPEND_LIVE@ bee.nvidia.mode=normal net.ifnames=0 biosdevname=0 mitigations=off transparent_hugepage=always numa_balancing=disable pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1 nowatchdog nosoftlockup
        initrd  @INITRD_LIVE@
    }

--- a/iso/builder/config/bootloaders/isolinux/live.cfg.in
+++ b/iso/builder/config/bootloaders/isolinux/live.cfg.in
@@ -3,31 +3,31 @@ label live-@FLAVOUR@-normal
    menu default
    linux @LINUX@
    initrd @INITRD@
-    append @APPEND_LIVE@ bee.nvidia.mode=normal
+    append @APPEND_LIVE@ bee.nvidia.mode=normal pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1

 label live-@FLAVOUR@-kms
    menu label EASY-BEE (^graphics/KMS)
    linux @LINUX@
    initrd @INITRD@
-    append @APPEND_LIVE@ bee.display=kms bee.nvidia.mode=normal
+    append @APPEND_LIVE@ bee.display=kms bee.nvidia.mode=normal pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1

 label live-@FLAVOUR@-toram
    menu label EASY-BEE (^load to RAM)
    linux @LINUX@
    initrd @INITRD@
-    append @APPEND_LIVE@ toram bee.nvidia.mode=normal
+    append @APPEND_LIVE@ toram bee.nvidia.mode=normal pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1

 label live-@FLAVOUR@-gsp-off
    menu label EASY-BEE (^NVIDIA GSP=off)
    linux @LINUX@
    initrd @INITRD@
-    append @APPEND_LIVE@ nomodeset bee.nvidia.mode=gsp-off
+    append @APPEND_LIVE@ nomodeset bee.nvidia.mode=gsp-off pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1

 label live-@FLAVOUR@-kms-gsp-off
    menu label EASY-BEE (g^raphics/KMS, GSP=off)
    linux @LINUX@
    initrd @INITRD@
-    append @APPEND_LIVE@ bee.display=kms bee.nvidia.mode=gsp-off
+    append @APPEND_LIVE@ bee.display=kms bee.nvidia.mode=gsp-off pcie_aspm=off intel_idle.max_cstate=1 processor.max_cstate=1

 label live-@FLAVOUR@-failsafe
    menu label EASY-BEE (^fail-safe)
--- a/iso/builder/config/hooks/normal/9000-bee-setup.hook.chroot
+++ b/iso/builder/config/hooks/normal/9000-bee-setup.hook.chroot
@@ -25,6 +25,7 @@ ensure_bee_console_user() {
 ensure_bee_console_user

 # Enable common bee services
+systemctl enable bee-hpc-tuning.service
 systemctl enable bee-network.service
 systemctl enable bee-preflight.service
 systemctl enable bee-audit.service
@@ -55,6 +56,7 @@ fi
 # nogpu: no GPU services needed

 # Ensure scripts are executable
+chmod +x /usr/local/bin/bee-hpc-tuning  2>/dev/null || true
 chmod +x /usr/local/bin/bee-network.sh  2>/dev/null || true
 chmod +x /usr/local/bin/bee-sshsetup   2>/dev/null || true
 chmod +x /usr/local/bin/bee-smoketest  2>/dev/null || true
--- a/iso/overlay/etc/systemd/system/bee-hpc-tuning.service
+++ b/iso/overlay/etc/systemd/system/bee-hpc-tuning.service
@@ -0,0 +1,14 @@
+[Unit]
+Description=Bee: HPC tuning (CPU governor, C-states)
+After=local-fs.target
+Before=bee-nvidia.service bee-audit.service
+
+[Service]
+Type=oneshot
+ExecStart=/usr/local/bin/bee-log-run /appdata/bee/export/bee-hpc-tuning.log /usr/local/bin/bee-hpc-tuning
+StandardOutput=journal
+StandardError=journal
+RemainAfterExit=yes
+
+[Install]
+WantedBy=multi-user.target
--- a/iso/overlay/usr/local/bin/bee-dcgmproftester-staggered
+++ b/iso/overlay/usr/local/bin/bee-dcgmproftester-staggered
@@ -0,0 +1,110 @@
+#!/bin/sh
+set -eu
+
+SECONDS=300
+STAGGER_SECONDS=180
+DEVICES=""
+EXCLUDE=""
+
+usage() {
+    echo "usage: $0 [--seconds N] [--stagger-seconds N] [--devices 0,1] [--exclude 2,3]" >&2
+    exit 2
+}
+
+normalize_list() {
+    echo "${1:-}" | tr ',' '\n' | sed 's/[[:space:]]//g' | awk 'NF' | sort -n | uniq | paste -sd, -
+}
+
+contains_csv() {
+    needle="$1"
+    haystack="${2:-}"
+    echo ",${haystack}," | grep -q ",${needle},"
+}
+
+resolve_dcgmproftester() {
+    for candidate in dcgmproftester dcgmproftester13 dcgmproftester12 dcgmproftester11; do
+        if command -v "${candidate}" >/dev/null 2>&1; then
+            command -v "${candidate}"
+            return 0
+        fi
+    done
+    return 1
+}
+
+while [ "$#" -gt 0 ]; do
+    case "$1" in
+        --seconds|-t) [ "$#" -ge 2 ] || usage; SECONDS="$2"; shift 2 ;;
+        --stagger-seconds) [ "$#" -ge 2 ] || usage; STAGGER_SECONDS="$2"; shift 2 ;;
+        --devices) [ "$#" -ge 2 ] || usage; DEVICES="$2"; shift 2 ;;
+        --exclude) [ "$#" -ge 2 ] || usage; EXCLUDE="$2"; shift 2 ;;
+        *) usage ;;
+    esac
+done
+
+PROF=$(resolve_dcgmproftester) || { echo "dcgmproftester not found in PATH" >&2; exit 1; }
+ALL_DEVICES=$(nvidia-smi --query-gpu=index --format=csv,noheader,nounits 2>/dev/null | sed 's/[[:space:]]//g' | awk 'NF' | paste -sd, -)
+[ -n "${ALL_DEVICES}" ] || { echo "nvidia-smi found no NVIDIA GPUs" >&2; exit 1; }
+
+DEVICES=$(normalize_list "${DEVICES}")
+EXCLUDE=$(normalize_list "${EXCLUDE}")
+SELECTED="${DEVICES}"
+if [ -z "${SELECTED}" ]; then
+    SELECTED="${ALL_DEVICES}"
+fi
+
+FINAL=""
+for id in $(echo "${SELECTED}" | tr ',' ' '); do
+    [ -n "${id}" ] || continue
+    if contains_csv "${id}" "${EXCLUDE}"; then
+        continue
+    fi
+    if [ -z "${FINAL}" ]; then
+        FINAL="${id}"
+    else
+        FINAL="${FINAL},${id}"
+    fi
+done
+
+[ -n "${FINAL}" ] || { echo "no NVIDIA GPUs selected after filters" >&2; exit 1; }
+
+echo "loader=dcgmproftester-staggered"
+echo "selected_gpus=${FINAL}"
+echo "stagger_seconds=${STAGGER_SECONDS}"
+
+TMP_DIR=$(mktemp -d)
+trap 'rm -rf "${TMP_DIR}"' EXIT INT TERM
+
+GPU_COUNT=$(echo "${FINAL}" | tr ',' '\n' | awk 'NF' | wc -l | tr -d '[:space:]')
+gpu_pos=0
+WORKERS=""
+for id in $(echo "${FINAL}" | tr ',' ' '); do
+    gpu_pos=$((gpu_pos + 1))
+    log="${TMP_DIR}/gpu-${id}.log"
+    extra_sec=$(( STAGGER_SECONDS * (GPU_COUNT - gpu_pos) ))
+    gpu_seconds=$(( SECONDS + extra_sec ))
+    echo "starting gpu ${id} seconds=${gpu_seconds}"
+    CUDA_VISIBLE_DEVICES="${id}" "${PROF}" --no-dcgm-validation -t 1004 -d "${gpu_seconds}" >"${log}" 2>&1 &
+    pid=$!
+    WORKERS="${WORKERS} ${pid}:${id}:${log}"
+    if [ "${STAGGER_SECONDS}" -gt 0 ] && [ "${gpu_pos}" -lt "${GPU_COUNT}" ]; then
+        sleep "${STAGGER_SECONDS}"
+    fi
+done
+
+status=0
+for spec in ${WORKERS}; do
+    pid=${spec%%:*}
+    rest=${spec#*:}
+    id=${rest%%:*}
+    log=${rest#*:}
+    if wait "${pid}"; then
+        echo "gpu ${id} finished: OK"
+    else
+        rc=$?
+        echo "gpu ${id} finished: FAILED rc=${rc}"
+        status=1
+    fi
+    sed "s/^/[gpu ${id}] /" "${log}" || true
+done
+
+exit "${status}"
--- a/iso/overlay/usr/local/bin/bee-gpu-burn
+++ b/iso/overlay/usr/local/bin/bee-gpu-burn
@@ -2,13 +2,14 @@
 set -eu

 SECONDS=5
+STAGGER_SECONDS=0
 SIZE_MB=0
 DEVICES=""
 EXCLUDE=""
 WORKER="/usr/local/lib/bee/bee-gpu-burn-worker"

 usage() {
-    echo "usage: $0 [--seconds N] [--size-mb N] [--devices 0,1] [--exclude 2,3]" >&2
+    echo "usage: $0 [--seconds N] [--stagger-seconds N] [--size-mb N] [--devices 0,1] [--exclude 2,3]" >&2
    exit 2
 }

@@ -25,6 +26,7 @@ contains_csv() {
 while [ "$#" -gt 0 ]; do
    case "$1" in
        --seconds|-t) [ "$#" -ge 2 ] || usage; SECONDS="$2"; shift 2 ;;
+        --stagger-seconds) [ "$#" -ge 2 ] || usage; STAGGER_SECONDS="$2"; shift 2 ;;
        --size-mb|-m) [ "$#" -ge 2 ] || usage; SIZE_MB="$2"; shift 2 ;;
        --devices) [ "$#" -ge 2 ] || usage; DEVICES="$2"; shift 2 ;;
        --exclude) [ "$#" -ge 2 ] || usage; EXCLUDE="$2"; shift 2 ;;
@@ -61,14 +63,18 @@ done

 echo "loader=bee-gpu-burn"
 echo "selected_gpus=${FINAL}"
+echo "stagger_seconds=${STAGGER_SECONDS}"

 export CUDA_DEVICE_ORDER="PCI_BUS_ID"

 TMP_DIR=$(mktemp -d)
 trap 'rm -rf "${TMP_DIR}"' EXIT INT TERM

+GPU_COUNT=$(echo "${FINAL}" | tr ',' '\n' | awk 'NF' | wc -l | tr -d '[:space:]')
+gpu_pos=0
 WORKERS=""
 for id in $(echo "${FINAL}" | tr ',' ' '); do
+    gpu_pos=$((gpu_pos + 1))
    log="${TMP_DIR}/gpu-${id}.log"
    gpu_size_mb="${SIZE_MB}"
    if [ "${gpu_size_mb}" -le 0 ] 2>/dev/null; then
@@ -79,11 +85,16 @@ for id in $(echo "${FINAL}" | tr ',' ' '); do
            gpu_size_mb=512
        fi
    fi
-    echo "starting gpu ${id} size=${gpu_size_mb}MB"
+    extra_sec=$(( STAGGER_SECONDS * (GPU_COUNT - gpu_pos) ))
+    gpu_seconds=$(( SECONDS + extra_sec ))
+    echo "starting gpu ${id} size=${gpu_size_mb}MB seconds=${gpu_seconds}"
    CUDA_VISIBLE_DEVICES="${id}" \
-        "${WORKER}" --device 0 --seconds "${SECONDS}" --size-mb "${gpu_size_mb}" >"${log}" 2>&1 &
+        "${WORKER}" --device 0 --seconds "${gpu_seconds}" --size-mb "${gpu_size_mb}" >"${log}" 2>&1 &
    pid=$!
    WORKERS="${WORKERS} ${pid}:${id}:${log}"
+    if [ "${STAGGER_SECONDS}" -gt 0 ] && [ "${gpu_pos}" -lt "${GPU_COUNT}" ]; then
+        sleep "${STAGGER_SECONDS}"
+    fi
 done

 status=0
--- a/iso/overlay/usr/local/bin/bee-hpc-tuning
+++ b/iso/overlay/usr/local/bin/bee-hpc-tuning
@@ -0,0 +1,41 @@
+#!/bin/sh
+# bee-hpc-tuning — apply HPC tuning for deterministic benchmarking
+# Called by bee-hpc-tuning.service at boot.
+
+log() { echo "[bee-hpc-tuning] $*"; }
+
+# ── CPU governor ────────────────────────────────────────────────────────────
+# Set all CPU cores to performance governor via sysfs.
+# cpupower is not available; write directly to scaling_governor.
+governor_ok=0
+governor_fail=0
+for gov_path in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do
+    [ -f "$gov_path" ] || continue
+    if echo performance > "$gov_path" 2>/dev/null; then
+        governor_ok=$((governor_ok + 1))
+    else
+        governor_fail=$((governor_fail + 1))
+    fi
+done
+
+if [ "$governor_ok" -gt 0 ] && [ "$governor_fail" -eq 0 ]; then
+    log "CPU governor set to performance on ${governor_ok} core(s)"
+elif [ "$governor_ok" -gt 0 ]; then
+    log "WARN: CPU governor: ${governor_ok} OK, ${governor_fail} failed"
+elif [ "$governor_fail" -gt 0 ]; then
+    log "WARN: failed to set CPU governor on ${governor_fail} core(s)"
+else
+    log "WARN: no cpufreq scaling_governor paths found (C-state governor or HW-controlled)"
+fi
+
+# ── Transparent Huge Pages ───────────────────────────────────────────────────
+# Kernel cmdline sets transparent_hugepage=always at boot, but confirm and log.
+thp_path=/sys/kernel/mm/transparent_hugepage/enabled
+if [ -f "$thp_path" ]; then
+    current=$(cat "$thp_path" 2>/dev/null)
+    log "transparent_hugepage: ${current}"
+else
+    log "WARN: transparent_hugepage sysfs path not found"
+fi
+
+log "done"
--- a/iso/overlay/usr/local/bin/bee-john-gpu-stress
+++ b/iso/overlay/usr/local/bin/bee-john-gpu-stress
@@ -2,6 +2,7 @@
 set -eu

 DURATION_SEC=300
+STAGGER_SECONDS=0
 DEVICES=""
 EXCLUDE=""
 FORMAT=""
@@ -12,7 +13,7 @@ export OCL_ICD_VENDORS="/etc/OpenCL/vendors"
 export LD_LIBRARY_PATH="/usr/lib:/usr/local/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"

 usage() {
-    echo "usage: $0 [--seconds N] [--devices 0,1] [--exclude 2,3] [--format name]" >&2
+    echo "usage: $0 [--seconds N] [--stagger-seconds N] [--devices 0,1] [--exclude 2,3] [--format name]" >&2
    exit 2
 }

@@ -118,6 +119,7 @@ ensure_opencl_ready() {
 while [ "$#" -gt 0 ]; do
    case "$1" in
        --seconds|-t) [ "$#" -ge 2 ] || usage; DURATION_SEC="$2"; shift 2 ;;
+        --stagger-seconds) [ "$#" -ge 2 ] || usage; STAGGER_SECONDS="$2"; shift 2 ;;
        --devices) [ "$#" -ge 2 ] || usage; DEVICES="$2"; shift 2 ;;
        --exclude) [ "$#" -ge 2 ] || usage; EXCLUDE="$2"; shift 2 ;;
        --format) [ "$#" -ge 2 ] || usage; FORMAT="$2"; shift 2 ;;
@@ -170,6 +172,7 @@ done
 echo "loader=john"
 echo "selected_gpus=${FINAL}"
 echo "john_devices=${JOHN_DEVICES}"
+echo "stagger_seconds=${STAGGER_SECONDS}"

 cd "${JOHN_DIR}"

@@ -232,14 +235,21 @@ trap cleanup EXIT INT TERM
 echo "format=${CHOSEN_FORMAT}"
 echo "target_seconds=${DURATION_SEC}"
 echo "slice_seconds=${TEST_SLICE_SECONDS}"
-DEADLINE=$(( $(date +%s) + DURATION_SEC ))
+TOTAL_DEVICES=$(echo "${JOHN_DEVICES}" | tr ',' '\n' | awk 'NF' | wc -l | tr -d '[:space:]')
 _first=1
+pos=0
 for opencl_id in $(echo "${JOHN_DEVICES}" | tr ',' ' '); do
+    pos=$((pos + 1))
    [ "${_first}" = "1" ] || sleep 3
    _first=0
-    run_john_loop "${opencl_id}" "${DEADLINE}" &
+    extra_sec=$(( STAGGER_SECONDS * (TOTAL_DEVICES - pos) ))
+    deadline=$(( $(date +%s) + DURATION_SEC + extra_sec ))
+    run_john_loop "${opencl_id}" "${deadline}" &
    pid=$!
    PIDS="${PIDS} ${pid}"
+    if [ "${STAGGER_SECONDS}" -gt 0 ] && [ "${pos}" -lt "${TOTAL_DEVICES}" ]; then
+        sleep "${STAGGER_SECONDS}"
+    fi
 done
 FAIL=0
 for pid in ${PIDS}; do
--- a/iso/overlay/usr/local/bin/bee-nvidia-load
+++ b/iso/overlay/usr/local/bin/bee-nvidia-load
@@ -21,8 +21,13 @@ read_nvidia_modules_flavor() {

 log "kernel: $(uname -r)"

-# Skip if no NVIDIA GPU present (PCI vendor 10de)
-if ! lspci -nn 2>/dev/null | grep -qi '10de:'; then
+# Skip if no NVIDIA display/compute GPU is present.
+# Match only display-class PCI functions (0300 VGA, 0302 3D controller) from vendor 10de.
+have_nvidia_gpu() {
+    lspci -Dn 2>/dev/null | awk '$2 ~ /^03(00|02):$/ && $3 ~ /^10de:/ { found=1; exit } END { exit(found ? 0 : 1) }'
+}
+
+if ! have_nvidia_gpu; then
    log "no NVIDIA GPU detected — skipping module load"
    exit 0
 fi
--- a/iso/overlay/usr/local/bin/bee-selfheal
+++ b/iso/overlay/usr/local/bin/bee-selfheal
@@ -14,7 +14,7 @@ log() {
 }

 have_nvidia_gpu() {
-    lspci -nn 2>/dev/null | grep -qi '10de:'
+    lspci -Dn 2>/dev/null | awk '$2 ~ /^03(00|02):$/ && $3 ~ /^10de:/ { found=1; exit } END { exit(found ? 0 : 1) }'
 }

 service_active() {
Author	SHA1	Message	Date
Mikhail Chusavitin	4f94ebcb2c	Add HPC tuning: PCIe ASPM off, C-states, performance CPU governor - grub.cfg + isolinux/live.cfg.in: add pcie_aspm=off, intel_idle.max_cstate=1 and processor.max_cstate=1 to all non-failsafe boot entries - bee-hpc-tuning: new script that sets all CPU cores to performance governor via sysfs and logs THP state at boot - bee-hpc-tuning.service: runs before bee-nvidia and bee-audit - 9000-bee-setup.hook.chroot: enable service and mark script executable Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-12 13:07:32 +03:00
Mikhail Chusavitin	05c1fde233	Warn on PCIe link speed degradation and collect lspci -vvv in techdump - collector/pcie: add applyPCIeLinkSpeedWarning that sets status=Warning and ErrorDescription when current link speed is below maximum negotiated speed (e.g. Gen1 running on a Gen5 slot) - collector/pcie: add pcieLinkSpeedRank helper for Gen string comparison - collector/pcie_filter_test: cover degraded and healthy link speed cases - platform/techdump: collect lspci -vvv → lspci-vvv.txt for LnkCap/LnkSta Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-12 12:42:17 +03:00
Michael Chus	825ef6b98a	Add USB export drive and LiveCD-in-RAM checks to Runtime Health - schema: add ToRAMStatus and USBExportPath fields to RuntimeHealth - platform/runtime.go: collectToRAMHealth (ok/warning/failed based on IsLiveMediaInRAM + toramActive) and collectUSBExportHealth (scans /proc/mounts + lsblk for writable USB-backed filesystems) - pages.go: add USB Export Drive and LiveCD in RAM rows to the health table Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-11 10:05:27 +03:00
Michael Chus	ba16021cdb	Fix GPU model propagation, export filenames, PSU/service status, and chart perf - nvidia.go: add Name field to nvidiaGPUInfo, include model name in nvidia-smi query, set dev.Model in enrichPCIeWithNVIDIAData - pages.go: fix duplicate GPU count in validate card summary (4 GPU: 4 x … → 4 x … GPU); fix PSU UNKNOWN fallback from hw.PowerSupplies; treat activating/deactivating/reloading service states as OK in Runtime Health - support_bundle.go: use "150405" time format (no colons) for exFAT compat - sat.go / benchmark.go / platform_stress.go / sat_fan_stress.go: remove .tar.gz archive creation from export dirs — export packs everything itself - charts_svg.go: add min-max downsampling (1400 pt cap) for SVG chart perf - benchmark_report.go / sat.go: normalize GPU fallback to "Unknown GPU" Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-11 10:05:27 +03:00
Mikhail Chusavitin	bb1218ddd4	Fix GPU inventory: exclude BMC virtual VGA, show real NVIDIA model names Two issues: 1. BMC/management VGA chips (e.g. Huawei iBMC Hi171x, ASPEED) were included in GPU inventory because shouldIncludePCIeDevice only checked the PCI class, not the device name. Added a name-based filter for known BMC/management patterns when the class is VGA/display/3d. 2. New NVIDIA GPUs (e.g. RTX PRO 6000 Blackwell, device ID 2bb5) showed as "Device 2bb5" because lspci's database lags behind. Added "name" to the nvidia-smi query and use it to override dev.Model during enrichment. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-10 13:57:26 +03:00
Mikhail Chusavitin	65faae8ede	Remove hpl from SAT run-all targets — no backend route exists hpl was listed in baseTargets and stressOnlyTargets but /api/sat/hpl/run was never registered, causing a 405 Method Not Allowed (not valid JSON) error when Validate one by one was triggered in stress mode. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-10 13:30:32 +03:00
Michael Chus	05241f2e0e	Redesign dashboard: split Runtime Health and Hardware Summary - Runtime Health now shows only LiveCD system status (services, tools, drivers, network, CUDA/ROCm) — hardware component rows removed - Hardware Summary now shows server components with readable descriptions (model, count×size) and component-status.json health badges - Add Network Adapters row to Hardware Summary - SFP module static info (vendor, PN, SN, connector, type, wavelength) now collected via ethtool -m regardless of carrier state - PSU statuses from IPMI audit written to component-status.json so PSU badge shows actual status after first audit instead of UNKNOWN Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-04-09 23:41:23 +03:00
Mikhail Chusavitin	c1690a084b	Fix app tests that mutate global defaults	2026-04-09 15:28:25 +03:00
Mikhail Chusavitin	9481ca2805	Add staged NVIDIA burn ramp-up mode	2026-04-09 15:21:14 +03:00
Mikhail Chusavitin	a78fdadd88	Refine validate and burn profile layout	2026-04-09 15:14:48 +03:00
Mikhail Chusavitin	4ef403898f	Tighten NVIDIA GPU PCI detection	2026-04-09 15:14:48 +03:00