Refine validate UI and runtime health table

audit/Makefile

@@ -1,9 +1,10 @@
 LISTEN ?= :8080
 AUDIT_PATH ?=
+EXPORT_DIR ?= $(CURDIR)/.tmp/export
 VERSION ?= $(shell sh ./scripts/resolve-version.sh)
 GO_LDFLAGS := -X main.Version=$(VERSION)
 
-RUN_ARGS := web --listen $(LISTEN)
+RUN_ARGS := web --listen $(LISTEN) --export-dir $(EXPORT_DIR)
 ifneq ($(AUDIT_PATH),)
 RUN_ARGS += --audit-path $(AUDIT_PATH)
 endif
@@ -11,6 +12,7 @@ endif
 .PHONY: run build test
 
 run:
+	mkdir -p $(EXPORT_DIR)
 	go run -ldflags "$(GO_LDFLAGS)" ./cmd/bee $(RUN_ARGS)
 
 build:

audit/cmd/bee/main.go

@@ -87,7 +87,7 @@ func printRootUsage(w io.Writer) {
   bee preflight --output stdout|file:<path>
   bee export  --target <device>
   bee support-bundle --output stdout|file:<path>
-  bee web     --listen :80 --audit-path `+app.DefaultAuditJSONPath+`
+  bee web     --listen :80 [--audit-path `+app.DefaultAuditJSONPath+`]
   bee sat nvidia|memory|storage|cpu [--duration <seconds>]
   bee benchmark nvidia [--profile standard|stability|overnight]
   bee version
@@ -296,7 +296,7 @@ func runWeb(args []string, stdout, stderr io.Writer) int {
 	fs := flag.NewFlagSet("web", flag.ContinueOnError)
 	fs.SetOutput(stderr)
 	listenAddr := fs.String("listen", ":8080", "listen address, e.g. :80")
-	auditPath := fs.String("audit-path", app.DefaultAuditJSONPath, "path to the latest audit JSON snapshot")
+	auditPath := fs.String("audit-path", "", "optional path to the latest audit JSON snapshot")
 	exportDir := fs.String("export-dir", app.DefaultExportDir, "directory with logs, SAT results, and support bundles")
 	title := fs.String("title", "Bee Hardware Audit", "page title")
 	fs.Usage = func() {

audit/internal/app/app.go

@@ -115,7 +115,12 @@ func (a *App) RunInstallToRAM(ctx context.Context, logFunc func(string)) error {
 type satRunner interface {
 	RunNvidiaAcceptancePack(baseDir string, logFunc func(string)) (string, error)
 	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)
+	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)
 	RunNvidiaStressPack(ctx context.Context, baseDir string, opts platform.NvidiaStressOptions, logFunc func(string)) (string, error)
 	RunMemoryAcceptancePack(ctx context.Context, baseDir string, logFunc func(string)) (string, error)
 	RunStorageAcceptancePack(ctx context.Context, baseDir string, logFunc func(string)) (string, error)
@@ -528,6 +533,13 @@ func (a *App) RunNvidiaAcceptancePackWithOptions(ctx context.Context, baseDir st
 	return ActionResult{Title: "NVIDIA DCGM", Body: body}, err
 }
 
+func (a *App) RunNvidiaTargetedStressValidatePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	if strings.TrimSpace(baseDir) == "" {
+		baseDir = DefaultSATBaseDir
+	}
+	return a.sat.RunNvidiaTargetedStressValidatePack(ctx, baseDir, durationSec, gpuIndices, logFunc)
+}
+
 func (a *App) RunNvidiaStressPack(baseDir string, opts platform.NvidiaStressOptions, logFunc func(string)) (string, error) {
 	return a.RunNvidiaStressPackCtx(context.Background(), baseDir, opts, logFunc)
 }
@@ -543,6 +555,34 @@ 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) {
+	if strings.TrimSpace(baseDir) == "" {
+		baseDir = DefaultSATBaseDir
+	}
+	return a.sat.RunNvidiaOfficialComputePack(ctx, baseDir, durationSec, gpuIndices, logFunc)
+}
+
+func (a *App) RunNvidiaTargetedPowerPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	if strings.TrimSpace(baseDir) == "" {
+		baseDir = DefaultSATBaseDir
+	}
+	return a.sat.RunNvidiaTargetedPowerPack(ctx, baseDir, durationSec, gpuIndices, logFunc)
+}
+
+func (a *App) RunNvidiaPulseTestPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	if strings.TrimSpace(baseDir) == "" {
+		baseDir = DefaultSATBaseDir
+	}
+	return a.sat.RunNvidiaPulseTestPack(ctx, baseDir, durationSec, gpuIndices, logFunc)
+}
+
+func (a *App) RunNvidiaBandwidthPack(ctx context.Context, baseDir string, gpuIndices []int, logFunc func(string)) (string, error) {
+	if strings.TrimSpace(baseDir) == "" {
+		baseDir = DefaultSATBaseDir
+	}
+	return a.sat.RunNvidiaBandwidthPack(ctx, baseDir, gpuIndices, logFunc)
+}
+
 func (a *App) RunNvidiaStressPackCtx(ctx context.Context, baseDir string, opts platform.NvidiaStressOptions, logFunc func(string)) (string, error) {
 	if strings.TrimSpace(baseDir) == "" {
 		baseDir = DefaultSATBaseDir
@@ -893,6 +933,12 @@ func latestSATSummaries() []string {
 		prefix string
 	}{
 		{label: "NVIDIA SAT", prefix: "gpu-nvidia-"},
+		{label: "NVIDIA Targeted Stress Validate (dcgmi diag targeted_stress)", prefix: "gpu-nvidia-targeted-stress-"},
+		{label: "NVIDIA Max Compute Load (dcgmproftester)", prefix: "gpu-nvidia-compute-"},
+		{label: "NVIDIA Targeted Power (dcgmi diag targeted_power)", prefix: "gpu-nvidia-targeted-power-"},
+		{label: "NVIDIA Pulse Test (dcgmi diag pulse_test)", prefix: "gpu-nvidia-pulse-"},
+		{label: "NVIDIA Interconnect Test (NCCL all_reduce_perf)", prefix: "gpu-nvidia-nccl-"},
+		{label: "NVIDIA Bandwidth Test (NVBandwidth)", prefix: "gpu-nvidia-bandwidth-"},
 		{label: "Memory SAT", prefix: "memory-"},
 		{label: "Storage SAT", prefix: "storage-"},
 		{label: "CPU SAT", prefix: "cpu-"},

audit/internal/app/app_test.go

@@ -120,16 +120,21 @@ func (f fakeTools) CheckTools(names []string) []platform.ToolStatus {
 }
 
 type fakeSAT struct {
-	runNvidiaFn          func(string) (string, error)
-	runNvidiaBenchmarkFn func(string, platform.NvidiaBenchmarkOptions) (string, error)
-	runNvidiaStressFn    func(string, platform.NvidiaStressOptions) (string, error)
-	runMemoryFn          func(string) (string, error)
-	runStorageFn         func(string) (string, error)
-	runCPUFn             func(string, int) (string, error)
-	detectVendorFn       func() string
-	listAMDGPUsFn        func() ([]platform.AMDGPUInfo, error)
-	runAMDPackFn         func(string) (string, error)
-	listNvidiaGPUsFn     func() ([]platform.NvidiaGPU, error)
+	runNvidiaFn               func(string) (string, error)
+	runNvidiaBenchmarkFn      func(string, platform.NvidiaBenchmarkOptions) (string, error)
+	runNvidiaStressFn         func(string, platform.NvidiaStressOptions) (string, error)
+	runNvidiaComputeFn        func(string, int, []int) (string, error)
+	runNvidiaPowerFn          func(string, int, []int) (string, error)
+	runNvidiaPulseFn          func(string, int, []int) (string, error)
+	runNvidiaBandwidthFn      func(string, []int) (string, error)
+	runNvidiaTargetedStressFn func(string, int, []int) (string, error)
+	runMemoryFn               func(string) (string, error)
+	runStorageFn              func(string) (string, error)
+	runCPUFn                  func(string, int) (string, error)
+	detectVendorFn            func() string
+	listAMDGPUsFn             func() ([]platform.AMDGPUInfo, error)
+	runAMDPackFn              func(string) (string, error)
+	listNvidiaGPUsFn          func() ([]platform.NvidiaGPU, error)
 }
 
 func (f fakeSAT) RunNvidiaAcceptancePack(baseDir string, _ func(string)) (string, error) {
@@ -147,6 +152,41 @@ func (f fakeSAT) RunNvidiaBenchmark(_ context.Context, baseDir string, opts plat
 	return f.runNvidiaFn(baseDir)
 }
 
+func (f fakeSAT) RunNvidiaTargetedStressValidatePack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ func(string)) (string, error) {
+	if f.runNvidiaTargetedStressFn != nil {
+		return f.runNvidiaTargetedStressFn(baseDir, durationSec, gpuIndices)
+	}
+	return f.runNvidiaFn(baseDir)
+}
+
+func (f fakeSAT) RunNvidiaOfficialComputePack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ func(string)) (string, error) {
+	if f.runNvidiaComputeFn != nil {
+		return f.runNvidiaComputeFn(baseDir, durationSec, gpuIndices)
+	}
+	return f.runNvidiaFn(baseDir)
+}
+
+func (f fakeSAT) RunNvidiaTargetedPowerPack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ func(string)) (string, error) {
+	if f.runNvidiaPowerFn != nil {
+		return f.runNvidiaPowerFn(baseDir, durationSec, gpuIndices)
+	}
+	return f.runNvidiaFn(baseDir)
+}
+
+func (f fakeSAT) RunNvidiaPulseTestPack(_ context.Context, baseDir string, durationSec int, gpuIndices []int, _ func(string)) (string, error) {
+	if f.runNvidiaPulseFn != nil {
+		return f.runNvidiaPulseFn(baseDir, durationSec, gpuIndices)
+	}
+	return f.runNvidiaFn(baseDir)
+}
+
+func (f fakeSAT) RunNvidiaBandwidthPack(_ context.Context, baseDir string, gpuIndices []int, _ func(string)) (string, error) {
+	if f.runNvidiaBandwidthFn != nil {
+		return f.runNvidiaBandwidthFn(baseDir, gpuIndices)
+	}
+	return f.runNvidiaFn(baseDir)
+}
+
 func (f fakeSAT) RunNvidiaStressPack(_ context.Context, baseDir string, opts platform.NvidiaStressOptions, _ func(string)) (string, error) {
 	if f.runNvidiaStressFn != nil {
 		return f.runNvidiaStressFn(baseDir, opts)

audit/internal/app/component_status_db.go

@@ -21,12 +21,12 @@ type ComponentStatusDB struct {
 
 // ComponentStatusRecord holds the current and historical health of one hardware component.
 type ComponentStatusRecord struct {
-	ComponentKey  string                  `json:"component_key"`
-	Status        string                  `json:"status"` // "OK", "Warning", "Critical", "Unknown"
-	LastCheckedAt time.Time               `json:"last_checked_at"`
-	LastChangedAt time.Time               `json:"last_changed_at"`
-	ErrorSummary  string                  `json:"error_summary,omitempty"`
-	History       []ComponentStatusEntry  `json:"history"`
+	ComponentKey  string                 `json:"component_key"`
+	Status        string                 `json:"status"` // "OK", "Warning", "Critical", "Unknown"
+	LastCheckedAt time.Time              `json:"last_checked_at"`
+	LastChangedAt time.Time              `json:"last_changed_at"`
+	ErrorSummary  string                 `json:"error_summary,omitempty"`
+	History       []ComponentStatusEntry `json:"history"`
 }
 
 // ComponentStatusEntry is one observation written to a component's history.
@@ -179,7 +179,9 @@ func ApplySATResultToDB(db *ComponentStatusDB, target, archivePath string) {
 
 	// Map SAT target to component keys.
 	switch target {
-	case "nvidia", "amd", "nvidia-stress", "amd-stress", "amd-mem", "amd-bandwidth":
+	case "nvidia", "nvidia-targeted-stress", "nvidia-compute", "nvidia-targeted-power", "nvidia-pulse",
+		"nvidia-interconnect", "nvidia-bandwidth", "amd", "nvidia-stress",
+		"amd-stress", "amd-mem", "amd-bandwidth":
 		db.Record("pcie:gpu:"+target, source, dbStatus, target+" SAT: "+overall)
 	case "memory", "memory-stress", "sat-stress":
 		db.Record("memory:all", source, dbStatus, target+" SAT: "+overall)

audit/internal/platform/runtime.go

@@ -135,12 +135,15 @@ func (s *System) runtimeToolStatuses(vendor string) []ToolStatus {
 	case "nvidia":
 		tools = append(tools, s.CheckTools([]string{
 			"nvidia-smi",
+			"dcgmi",
+			"nv-hostengine",
 			"nvidia-bug-report.sh",
 			"bee-gpu-burn",
 			"bee-john-gpu-stress",
 			"bee-nccl-gpu-stress",
 			"all_reduce_perf",
 		})...)
+		tools = append(tools, resolvedToolStatus("dcgmproftester", dcgmProfTesterCandidates...))
 	case "amd":
 		tool := ToolStatus{Name: "rocm-smi"}
 		if cmd, err := resolveROCmSMICommand(); err == nil && len(cmd) > 0 {
@@ -155,6 +158,16 @@ func (s *System) runtimeToolStatuses(vendor string) []ToolStatus {
 	return tools
 }
 
+func resolvedToolStatus(display string, candidates ...string) ToolStatus {
+	for _, candidate := range candidates {
+		path, err := exec.LookPath(candidate)
+		if err == nil {
+			return ToolStatus{Name: display, Path: path, OK: true}
+		}
+	}
+	return ToolStatus{Name: display}
+}
+
 func (s *System) collectGPURuntimeHealth(vendor string, health *schema.RuntimeHealth) {
 	lsmodText := commandText("lsmod")
 

audit/internal/platform/sat.go

@@ -38,6 +38,12 @@ var (
 		"/opt/rocm/bin/rvs",
 		"/opt/rocm-*/bin/rvs",
 	}
+	dcgmProfTesterCandidates = []string{
+		"dcgmproftester",
+		"dcgmproftester13",
+		"dcgmproftester12",
+		"dcgmproftester11",
+	}
 )
 
 // streamExecOutput runs cmd and streams each output line to logFunc (if non-nil).
@@ -277,6 +283,80 @@ 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) {
+	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
+	}
+	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-compute", []satJob{
+		{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
+		{name: "02-dcgmi-version.log", cmd: []string{"dcgmi", "-v"}},
+		{
+			name:       "03-dcgmproftester.log",
+			cmd:        profCmd,
+			env:        nvidiaVisibleDevicesEnv(selected),
+			collectGPU: true,
+			gpuIndices: selected,
+		},
+		{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)
+}
+
+func (s *System) RunNvidiaTargetedPowerPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	selected, err := resolveDCGMGPUIndices(gpuIndices)
+	if err != nil {
+		return "", err
+	}
+	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-targeted-power", []satJob{
+		{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
+		{
+			name:       "02-dcgmi-targeted-power.log",
+			cmd:        nvidiaDCGMNamedDiagCommand("targeted_power", normalizeNvidiaBurnDuration(durationSec), selected),
+			collectGPU: true,
+			gpuIndices: selected,
+		},
+		{name: "03-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)
+}
+
+func (s *System) RunNvidiaPulseTestPack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	selected, err := resolveDCGMGPUIndices(gpuIndices)
+	if err != nil {
+		return "", err
+	}
+	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-pulse", []satJob{
+		{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
+		{
+			name:       "02-dcgmi-pulse-test.log",
+			cmd:        nvidiaDCGMNamedDiagCommand("pulse_test", normalizeNvidiaBurnDuration(durationSec), selected),
+			collectGPU: true,
+			gpuIndices: selected,
+		},
+		{name: "03-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)
+}
+
+func (s *System) RunNvidiaBandwidthPack(ctx context.Context, baseDir string, gpuIndices []int, logFunc func(string)) (string, error) {
+	selected, err := resolveDCGMGPUIndices(gpuIndices)
+	if err != nil {
+		return "", err
+	}
+	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-bandwidth", []satJob{
+		{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
+		{
+			name:       "02-dcgmi-nvbandwidth.log",
+			cmd:        nvidiaDCGMNamedDiagCommand("nvbandwidth", 0, selected),
+			collectGPU: true,
+			gpuIndices: selected,
+		},
+		{name: "03-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)
+}
+
 func (s *System) RunNvidiaAcceptancePack(baseDir string, logFunc func(string)) (string, error) {
 	return runAcceptancePackCtx(context.Background(), baseDir, "gpu-nvidia", nvidiaSATJobs(), logFunc)
 }
@@ -293,6 +373,23 @@ func (s *System) RunNvidiaAcceptancePackWithOptions(ctx context.Context, baseDir
 	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia", nvidiaDCGMJobs(diagLevel, resolvedGPUIndices), logFunc)
 }
 
+func (s *System) RunNvidiaTargetedStressValidatePack(ctx context.Context, baseDir string, durationSec int, gpuIndices []int, logFunc func(string)) (string, error) {
+	selected, err := resolveDCGMGPUIndices(gpuIndices)
+	if err != nil {
+		return "", err
+	}
+	return runAcceptancePackCtx(ctx, baseDir, "gpu-nvidia-targeted-stress", []satJob{
+		{name: "01-nvidia-smi-q.log", cmd: []string{"nvidia-smi", "-q"}},
+		{
+			name:       "02-dcgmi-targeted-stress.log",
+			cmd:        nvidiaDCGMNamedDiagCommand("targeted_stress", normalizeNvidiaBurnDuration(durationSec), selected),
+			collectGPU: true,
+			gpuIndices: selected,
+		},
+		{name: "03-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)
+}
+
 func resolveDCGMGPUIndices(gpuIndices []int) ([]int, error) {
 	if len(gpuIndices) > 0 {
 		return dedupeSortedIndices(gpuIndices), nil
@@ -473,6 +570,31 @@ func nvidiaDCGMJobs(diagLevel int, gpuIndices []int) []satJob {
 	}
 }
 
+func nvidiaDCGMNamedDiagCommand(name string, durationSec int, gpuIndices []int) []string {
+	args := []string{"dcgmi", "diag", "-r", name}
+	if durationSec > 0 {
+		args = append(args, "-p", fmt.Sprintf("%s.test_duration=%d", name, durationSec))
+	}
+	if len(gpuIndices) > 0 {
+		args = append(args, "-i", joinIndexList(gpuIndices))
+	}
+	return args
+}
+
+func normalizeNvidiaBurnDuration(durationSec int) int {
+	if durationSec <= 0 {
+		return 300
+	}
+	return durationSec
+}
+
+func nvidiaVisibleDevicesEnv(gpuIndices []int) []string {
+	if len(gpuIndices) == 0 {
+		return nil
+	}
+	return []string{"CUDA_VISIBLE_DEVICES=" + joinIndexList(gpuIndices)}
+}
+
 func runAcceptancePackCtx(ctx context.Context, baseDir, prefix string, jobs []satJob, logFunc func(string)) (string, error) {
 	if ctx == nil {
 		ctx = context.Background()
@@ -642,6 +764,7 @@ func classifySATResult(name string, out []byte, err error) (string, int) {
 	}
 	if strings.Contains(text, "unsupported") ||
 		strings.Contains(text, "not supported") ||
+		strings.Contains(text, "not found in path") ||
 		strings.Contains(text, "invalid opcode") ||
 		strings.Contains(text, "unknown command") ||
 		strings.Contains(text, "not implemented") ||
@@ -748,6 +871,15 @@ func resolveROCmSMICommand(args ...string) ([]string, error) {
 	return nil, errors.New("rocm-smi not found in PATH or under /opt/rocm")
 }
 
+func resolveDCGMProfTesterCommand(args ...string) ([]string, error) {
+	for _, candidate := range dcgmProfTesterCandidates {
+		if path, err := satLookPath(candidate); err == nil {
+			return append([]string{path}, args...), nil
+		}
+	}
+	return nil, errors.New("dcgmproftester not found in PATH")
+}
+
 func ensureAMDRuntimeReady() error {
 	if _, err := os.Stat("/dev/kfd"); err == nil {
 		return nil

audit/internal/platform/sat_test.go

@@ -195,6 +195,53 @@ func TestResolveDCGMGPUIndicesKeepsExplicitSelection(t *testing.T) {
 	}
 }
 
+func TestResolveDCGMProfTesterCommandUsesVersionedBinary(t *testing.T) {
+	oldLookPath := satLookPath
+	satLookPath = func(file string) (string, error) {
+		switch file {
+		case "dcgmproftester13":
+			return "/usr/bin/dcgmproftester13", nil
+		default:
+			return "", exec.ErrNotFound
+		}
+	}
+	t.Cleanup(func() { satLookPath = oldLookPath })
+
+	cmd, err := resolveDCGMProfTesterCommand("--no-dcgm-validation", "-t", "1004")
+	if err != nil {
+		t.Fatalf("resolveDCGMProfTesterCommand error: %v", err)
+	}
+	if len(cmd) != 4 {
+		t.Fatalf("cmd len=%d want 4 (%v)", len(cmd), cmd)
+	}
+	if cmd[0] != "/usr/bin/dcgmproftester13" {
+		t.Fatalf("cmd[0]=%q want /usr/bin/dcgmproftester13", cmd[0])
+	}
+}
+
+func TestNvidiaDCGMNamedDiagCommandUsesDurationAndSelection(t *testing.T) {
+	cmd := nvidiaDCGMNamedDiagCommand("targeted_power", 900, []int{3, 1})
+	want := []string{"dcgmi", "diag", "-r", "targeted_power", "-p", "targeted_power.test_duration=900", "-i", "3,1"}
+	if len(cmd) != len(want) {
+		t.Fatalf("cmd len=%d want %d (%v)", len(cmd), len(want), cmd)
+	}
+	for i := range want {
+		if cmd[i] != want[i] {
+			t.Fatalf("cmd[%d]=%q want %q", i, cmd[i], want[i])
+		}
+	}
+}
+
+func TestNvidiaVisibleDevicesEnvUsesSelectedGPUs(t *testing.T) {
+	env := nvidiaVisibleDevicesEnv([]int{0, 2, 4})
+	if len(env) != 1 {
+		t.Fatalf("env len=%d want 1 (%v)", len(env), env)
+	}
+	if env[0] != "CUDA_VISIBLE_DEVICES=0,2,4" {
+		t.Fatalf("env[0]=%q want CUDA_VISIBLE_DEVICES=0,2,4", env[0])
+	}
+}
+
 func TestNvidiaStressArchivePrefixByLoader(t *testing.T) {
 	t.Parallel()
 

audit/internal/webui/api.go

@@ -581,15 +581,32 @@ func (h *handler) handleAPIGPUTools(w http.ResponseWriter, _ *http.Request) {
 	nvidiaUp := nvidiaErr == nil
 	amdUp := amdErr == nil
 	_, dcgmErr := exec.LookPath("dcgmi")
+	_, ncclStressErr := exec.LookPath("bee-nccl-gpu-stress")
+	_, johnErr := exec.LookPath("bee-john-gpu-stress")
+	_, beeBurnErr := exec.LookPath("bee-gpu-burn")
+	_, nvBandwidthErr := exec.LookPath("nvbandwidth")
+	profErr := lookPathAny("dcgmproftester", "dcgmproftester13", "dcgmproftester12", "dcgmproftester11")
 	writeJSON(w, []toolEntry{
-		{ID: "bee-gpu-burn", Available: nvidiaUp, Vendor: "nvidia"},
-		{ID: "dcgm", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
-		{ID: "john", Available: nvidiaUp, Vendor: "nvidia"},
-		{ID: "nccl", Available: nvidiaUp, Vendor: "nvidia"},
+		{ID: "nvidia-compute", Available: nvidiaUp && profErr == nil, Vendor: "nvidia"},
+		{ID: "nvidia-targeted-power", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
+		{ID: "nvidia-pulse", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
+		{ID: "nvidia-interconnect", Available: nvidiaUp && ncclStressErr == nil, Vendor: "nvidia"},
+		{ID: "nvidia-bandwidth", Available: nvidiaUp && dcgmErr == nil && nvBandwidthErr == nil, Vendor: "nvidia"},
+		{ID: "bee-gpu-burn", Available: nvidiaUp && beeBurnErr == nil, Vendor: "nvidia"},
+		{ID: "john", Available: nvidiaUp && johnErr == nil, Vendor: "nvidia"},
 		{ID: "rvs", Available: amdUp, Vendor: "amd"},
 	})
 }
 
+func lookPathAny(names ...string) error {
+	for _, name := range names {
+		if _, err := exec.LookPath(name); err == nil {
+			return nil
+		}
+	}
+	return exec.ErrNotFound
+}
+
 // ── System ────────────────────────────────────────────────────────────────────
 
 func (h *handler) handleAPIRAMStatus(w http.ResponseWriter, r *http.Request) {
@@ -628,7 +645,7 @@ func (h *handler) handleAPIInstallToRAM(w http.ResponseWriter, r *http.Request)
 
 var standardTools = []string{
 	"dmidecode", "smartctl", "nvme", "lspci", "ipmitool",
-	"nvidia-smi", "memtester", "stress-ng", "nvtop",
+	"nvidia-smi", "dcgmi", "nv-hostengine", "memtester", "stress-ng", "nvtop",
 	"mstflint", "qrencode",
 }
 

audit/internal/webui/kmsg_watcher.go

@@ -232,7 +232,8 @@ func truncate(s string, max int) string {
 // isSATTarget returns true for task targets that run hardware acceptance tests.
 func isSATTarget(target string) bool {
 	switch target {
-	case "nvidia", "nvidia-benchmark", "nvidia-stress", "memory", "memory-stress", "storage",
+	case "nvidia", "nvidia-targeted-stress", "nvidia-benchmark", "nvidia-compute", "nvidia-targeted-power", "nvidia-pulse",
+		"nvidia-interconnect", "nvidia-bandwidth", "nvidia-stress", "memory", "memory-stress", "storage",
 		"cpu", "sat-stress", "amd", "amd-mem", "amd-bandwidth", "amd-stress",
 		"platform-stress":
 		return true

audit/internal/webui/server.go

@@ -237,6 +237,12 @@ func NewHandler(opts HandlerOptions) http.Handler {
 
 	// SAT
 	mux.HandleFunc("POST /api/sat/nvidia/run", h.handleAPISATRun("nvidia"))
+	mux.HandleFunc("POST /api/sat/nvidia-targeted-stress/run", h.handleAPISATRun("nvidia-targeted-stress"))
+	mux.HandleFunc("POST /api/sat/nvidia-compute/run", h.handleAPISATRun("nvidia-compute"))
+	mux.HandleFunc("POST /api/sat/nvidia-targeted-power/run", h.handleAPISATRun("nvidia-targeted-power"))
+	mux.HandleFunc("POST /api/sat/nvidia-pulse/run", h.handleAPISATRun("nvidia-pulse"))
+	mux.HandleFunc("POST /api/sat/nvidia-interconnect/run", h.handleAPISATRun("nvidia-interconnect"))
+	mux.HandleFunc("POST /api/sat/nvidia-bandwidth/run", h.handleAPISATRun("nvidia-bandwidth"))
 	mux.HandleFunc("POST /api/sat/nvidia-stress/run", h.handleAPISATRun("nvidia-stress"))
 	mux.HandleFunc("POST /api/sat/memory/run", h.handleAPISATRun("memory"))
 	mux.HandleFunc("POST /api/sat/storage/run", h.handleAPISATRun("storage"))
@@ -1182,6 +1188,11 @@ func (h *handler) handleAPIMetricsExportCSV(w http.ResponseWriter, r *http.Reque
 
 func (h *handler) handleReady(w http.ResponseWriter, r *http.Request) {
 	w.Header().Set("Cache-Control", "no-store")
+	if strings.TrimSpace(h.opts.AuditPath) == "" {
+		w.WriteHeader(http.StatusOK)
+		_, _ = w.Write([]byte("ready"))
+		return
+	}
 	if _, err := os.Stat(h.opts.AuditPath); err != nil {
 		w.WriteHeader(http.StatusServiceUnavailable)
 		_, _ = w.Write([]byte("starting"))

audit/internal/webui/server_test.go

@@ -543,7 +543,7 @@ func TestRootShowsRunAuditButtonWhenSnapshotMissing(t *testing.T) {
 		t.Fatalf("status=%d", rec.Code)
 	}
 	body := rec.Body.String()
-	if !strings.Contains(body, `Run Audit`) {
+	if !strings.Contains(body, `onclick="auditModalRun()">Run audit</button>`) {
 		t.Fatalf("dashboard missing run audit button: %s", body)
 	}
 	if strings.Contains(body, `No audit data`) {
@@ -551,6 +551,18 @@ func TestRootShowsRunAuditButtonWhenSnapshotMissing(t *testing.T) {
 	}
 }
 
+func TestReadyIsOKWhenAuditPathIsUnset(t *testing.T) {
+	handler := NewHandler(HandlerOptions{})
+	rec := httptest.NewRecorder()
+	handler.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/api/ready", nil))
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rec.Code, rec.Body.String())
+	}
+	if strings.TrimSpace(rec.Body.String()) != "ready" {
+		t.Fatalf("body=%q want ready", rec.Body.String())
+	}
+}
+
 func TestAuditPageRendersViewerFrameAndActions(t *testing.T) {
 	dir := t.TempDir()
 	path := filepath.Join(dir, "audit.json")
@@ -638,7 +650,27 @@ func TestBenchmarkPageRendersGPUSelectionControls(t *testing.T) {
 	}
 }
 
-func TestBurnPageRendersOfficialNVIDIADCGMAndNCCLInterconnectLabel(t *testing.T) {
+func TestValidatePageRendersNvidiaTargetedStressCard(t *testing.T) {
+	handler := NewHandler(HandlerOptions{})
+	rec := httptest.NewRecorder()
+	handler.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/validate", nil))
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status=%d", rec.Code)
+	}
+	body := rec.Body.String()
+	for _, needle := range []string{
+		`NVIDIA GPU Targeted Stress`,
+		`nvidia-targeted-stress`,
+		`controlled NVIDIA DCGM load`,
+		`<code>dcgmi diag targeted_stress</code>`,
+	} {
+		if !strings.Contains(body, needle) {
+			t.Fatalf("validate page missing %q: %s", needle, body)
+		}
+	}
+}
+
+func TestBurnPageRendersGoalBasedNVIDIACards(t *testing.T) {
 	handler := NewHandler(HandlerOptions{})
 	rec := httptest.NewRecorder()
 	handler.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/burn", nil))
@@ -647,10 +679,11 @@ func TestBurnPageRendersOfficialNVIDIADCGMAndNCCLInterconnectLabel(t *testing.T)
 	}
 	body := rec.Body.String()
 	for _, needle := range []string{
-		`DCGM Diagnostics (Official NVIDIA)`,
-		`NCCL all_reduce_perf (Interconnect)`,
-		`DCGM is the official NVIDIA diagnostic path`,
-		`burn-gpu-dcgm`,
+		`NVIDIA Max Compute Load`,
+		`dcgmproftester`,
+		`targeted_stress remain in <a href="/validate">Validate</a>`,
+		`NVIDIA Interconnect Test (NCCL all_reduce_perf)`,
+		`id="burn-gpu-list"`,
 	} {
 		if !strings.Contains(body, needle) {
 			t.Fatalf("burn page missing %q: %s", needle, body)
@@ -813,3 +846,98 @@ func TestRuntimeHealthEndpointReturnsJSON(t *testing.T) {
 		t.Fatalf("body=%q want %q", strings.TrimSpace(rec.Body.String()), body)
 	}
 }
+
+func TestDashboardRendersRuntimeHealthTable(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.json")
+	exportDir := filepath.Join(dir, "export")
+	if err := os.MkdirAll(exportDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(`{"collected_at":"2026-03-15T00:00:00Z","hardware":{"board":{"serial_number":"SERIAL-1"}}}`), 0644); err != nil {
+		t.Fatal(err)
+	}
+	health := `{
+  "status":"PARTIAL",
+  "checked_at":"2026-03-16T10:00:00Z",
+  "export_dir":"/tmp/export",
+  "driver_ready":true,
+  "cuda_ready":false,
+  "network_status":"PARTIAL",
+  "issues":[
+    {"code":"dhcp_partial","description":"At least one interface did not obtain IPv4 connectivity."},
+    {"code":"cuda_runtime_not_ready","description":"CUDA runtime is not ready for GPU SAT."}
+  ],
+  "tools":[
+    {"name":"dmidecode","ok":true},
+    {"name":"nvidia-smi","ok":false}
+  ],
+  "services":[
+    {"name":"bee-web","status":"active"},
+    {"name":"bee-nvidia","status":"inactive"}
+  ]
+}`
+	if err := os.WriteFile(filepath.Join(exportDir, "runtime-health.json"), []byte(health), 0644); err != nil {
+		t.Fatal(err)
+	}
+	componentStatus := `[
+  {
+    "component_key":"cpu:all",
+    "status":"Warning",
+    "error_summary":"cpu SAT: FAILED",
+    "history":[{"at":"2026-03-16T10:00:00Z","status":"Warning","source":"sat:cpu","detail":"cpu SAT: FAILED"}]
+  },
+  {
+    "component_key":"memory:all",
+    "status":"OK",
+    "history":[{"at":"2026-03-16T10:01:00Z","status":"OK","source":"sat:memory","detail":"memory SAT: OK"}]
+  },
+  {
+    "component_key":"storage:nvme0n1",
+    "status":"Critical",
+    "error_summary":"storage SAT: FAILED",
+    "history":[{"at":"2026-03-16T10:02:00Z","status":"Critical","source":"sat:storage","detail":"storage SAT: FAILED"}]
+  },
+  {
+    "component_key":"pcie:gpu:nvidia",
+    "status":"Warning",
+    "error_summary":"nvidia SAT: FAILED",
+    "history":[{"at":"2026-03-16T10:03:00Z","status":"Warning","source":"sat:nvidia","detail":"nvidia SAT: FAILED"}]
+  }
+]`
+	if err := os.WriteFile(filepath.Join(exportDir, "component-status.json"), []byte(componentStatus), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	handler := NewHandler(HandlerOptions{AuditPath: path, ExportDir: exportDir})
+	rec := httptest.NewRecorder()
+	handler.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/", nil))
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rec.Code, rec.Body.String())
+	}
+	body := rec.Body.String()
+	for _, needle := range []string{
+		`Runtime Health`,
+		`<th>Check</th><th>Status</th><th>Source</th><th>Issue</th>`,
+		`Export Directory`,
+		`Network`,
+		`NVIDIA/AMD Driver`,
+		`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`,
+	} {
+		if !strings.Contains(body, needle) {
+			t.Fatalf("dashboard missing %q: %s", needle, body)
+		}
+	}
+}

audit/internal/webui/tasks.go

@@ -30,23 +30,29 @@ const (
 
 // taskNames maps target → human-readable name for validate (SAT) runs.
 var taskNames = map[string]string{
-	"nvidia":           "NVIDIA SAT",
-	"nvidia-benchmark": "NVIDIA Benchmark",
-	"nvidia-stress":    "NVIDIA GPU Stress",
-	"memory":           "Memory SAT",
-	"storage":          "Storage SAT",
-	"cpu":              "CPU SAT",
-	"amd":              "AMD GPU SAT",
-	"amd-mem":          "AMD GPU MEM Integrity",
-	"amd-bandwidth":    "AMD GPU MEM Bandwidth",
-	"amd-stress":       "AMD GPU Burn-in",
-	"memory-stress":    "Memory Burn-in",
-	"sat-stress":       "SAT Stress (stressapptest)",
-	"platform-stress":  "Platform Thermal Cycling",
-	"audit":            "Audit",
-	"support-bundle":   "Support Bundle",
-	"install":          "Install to Disk",
-	"install-to-ram":   "Install to RAM",
+	"nvidia":                 "NVIDIA SAT",
+	"nvidia-targeted-stress": "NVIDIA Targeted Stress Validate (dcgmi diag targeted_stress)",
+	"nvidia-benchmark":       "NVIDIA Benchmark",
+	"nvidia-compute":         "NVIDIA Max Compute Load (dcgmproftester)",
+	"nvidia-targeted-power":  "NVIDIA Targeted Power (dcgmi diag targeted_power)",
+	"nvidia-pulse":           "NVIDIA Pulse Test (dcgmi diag pulse_test)",
+	"nvidia-interconnect":    "NVIDIA Interconnect Test (NCCL all_reduce_perf)",
+	"nvidia-bandwidth":       "NVIDIA Bandwidth Test (NVBandwidth)",
+	"nvidia-stress":          "NVIDIA GPU Stress",
+	"memory":                 "Memory SAT",
+	"storage":                "Storage SAT",
+	"cpu":                    "CPU SAT",
+	"amd":                    "AMD GPU SAT",
+	"amd-mem":                "AMD GPU MEM Integrity",
+	"amd-bandwidth":          "AMD GPU MEM Bandwidth",
+	"amd-stress":             "AMD GPU Burn-in",
+	"memory-stress":          "Memory Burn-in",
+	"sat-stress":             "SAT Stress (stressapptest)",
+	"platform-stress":        "Platform Thermal Cycling",
+	"audit":                  "Audit",
+	"support-bundle":         "Support Bundle",
+	"install":                "Install to Disk",
+	"install-to-ram":         "Install to RAM",
 }
 
 // burnNames maps target → human-readable name when a burn profile is set.
@@ -134,45 +140,53 @@ type persistedTask struct {
 }
 
 type burnPreset struct {
-	NvidiaDiag  int
 	DurationSec int
 }
 
 func resolveBurnPreset(profile string) burnPreset {
 	switch profile {
 	case "overnight":
-		return burnPreset{NvidiaDiag: 4, DurationSec: 8 * 60 * 60}
+		return burnPreset{DurationSec: 8 * 60 * 60}
 	case "acceptance":
-		return burnPreset{NvidiaDiag: 3, DurationSec: 60 * 60}
+		return burnPreset{DurationSec: 60 * 60}
 	default:
-		return burnPreset{NvidiaDiag: 1, DurationSec: 5 * 60}
+		return burnPreset{DurationSec: 5 * 60}
 	}
 }
 
 func resolvePlatformStressPreset(profile string) platform.PlatformStressOptions {
+	acceptanceCycles := []platform.PlatformStressCycle{
+		{LoadSec: 85, IdleSec: 5},
+		{LoadSec: 80, IdleSec: 10},
+		{LoadSec: 55, IdleSec: 5},
+		{LoadSec: 60, IdleSec: 0},
+		{LoadSec: 100, IdleSec: 10},
+		{LoadSec: 145, IdleSec: 15},
+		{LoadSec: 190, IdleSec: 20},
+		{LoadSec: 235, IdleSec: 25},
+		{LoadSec: 280, IdleSec: 30},
+		{LoadSec: 325, IdleSec: 35},
+		{LoadSec: 370, IdleSec: 40},
+		{LoadSec: 415, IdleSec: 45},
+		{LoadSec: 460, IdleSec: 50},
+		{LoadSec: 510, IdleSec: 0},
+	}
+
 	switch profile {
 	case "overnight":
-		return platform.PlatformStressOptions{Cycles: []platform.PlatformStressCycle{
-			{LoadSec: 600, IdleSec: 120},
-			{LoadSec: 600, IdleSec: 60},
-			{LoadSec: 600, IdleSec: 30},
-			{LoadSec: 600, IdleSec: 120},
-			{LoadSec: 600, IdleSec: 60},
-			{LoadSec: 600, IdleSec: 30},
-			{LoadSec: 600, IdleSec: 120},
-			{LoadSec: 600, IdleSec: 60},
-		}}
+		cycles := make([]platform.PlatformStressCycle, 0, len(acceptanceCycles)*8)
+		for range 8 {
+			cycles = append(cycles, acceptanceCycles...)
+		}
+		return platform.PlatformStressOptions{Cycles: cycles}
 	case "acceptance":
-		return platform.PlatformStressOptions{Cycles: []platform.PlatformStressCycle{
-			{LoadSec: 300, IdleSec: 60},
-			{LoadSec: 300, IdleSec: 30},
-			{LoadSec: 300, IdleSec: 60},
-			{LoadSec: 300, IdleSec: 30},
-		}}
+		return platform.PlatformStressOptions{Cycles: acceptanceCycles}
 	default: // smoke
 		return platform.PlatformStressOptions{Cycles: []platform.PlatformStressCycle{
-			{LoadSec: 90, IdleSec: 60},
-			{LoadSec: 90, IdleSec: 30},
+			{LoadSec: 85, IdleSec: 5},
+			{LoadSec: 80, IdleSec: 10},
+			{LoadSec: 55, IdleSec: 5},
+			{LoadSec: 60, IdleSec: 0},
 		}}
 	}
 }
@@ -536,9 +550,6 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 			break
 		}
 		diagLevel := t.params.DiagLevel
-		if t.params.BurnProfile != "" && diagLevel <= 0 {
-			diagLevel = resolveBurnPreset(t.params.BurnProfile).NvidiaDiag
-		}
 		if len(t.params.GPUIndices) > 0 || diagLevel > 0 {
 			result, e := a.RunNvidiaAcceptancePackWithOptions(
 				ctx, "", diagLevel, t.params.GPUIndices, j.append,
@@ -551,6 +562,16 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 		} else {
 			archive, err = a.RunNvidiaAcceptancePack("", j.append)
 		}
+	case "nvidia-targeted-stress":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		dur := t.params.Duration
+		if dur <= 0 {
+			dur = 300
+		}
+		archive, err = a.RunNvidiaTargetedStressValidatePack(ctx, "", dur, t.params.GPUIndices, j.append)
 	case "nvidia-benchmark":
 		if a == nil {
 			err = fmt.Errorf("app not configured")
@@ -563,6 +584,56 @@ func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
 			ExcludeGPUIndices: t.params.ExcludeGPUIndices,
 			RunNCCL:           t.params.RunNCCL,
 		}, j.append)
+	case "nvidia-compute":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		dur := t.params.Duration
+		if t.params.BurnProfile != "" && dur <= 0 {
+			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
+		}
+		archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, j.append)
+	case "nvidia-targeted-power":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		dur := t.params.Duration
+		if t.params.BurnProfile != "" && dur <= 0 {
+			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
+		}
+		archive, err = a.RunNvidiaTargetedPowerPack(ctx, "", dur, t.params.GPUIndices, j.append)
+	case "nvidia-pulse":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		dur := t.params.Duration
+		if t.params.BurnProfile != "" && dur <= 0 {
+			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
+		}
+		archive, err = a.RunNvidiaPulseTestPack(ctx, "", dur, t.params.GPUIndices, j.append)
+	case "nvidia-bandwidth":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		archive, err = a.RunNvidiaBandwidthPack(ctx, "", t.params.GPUIndices, j.append)
+	case "nvidia-interconnect":
+		if a == nil {
+			err = fmt.Errorf("app not configured")
+			break
+		}
+		dur := t.params.Duration
+		if t.params.BurnProfile != "" && dur <= 0 {
+			dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
+		}
+		archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
+			DurationSec: dur,
+			Loader:      platform.NvidiaStressLoaderNCCL,
+			GPUIndices:  t.params.GPUIndices,
+		}, j.append)
 	case "nvidia-stress":
 		if a == nil {
 			err = fmt.Errorf("app not configured")

audit/internal/webui/tasks_test.go

@@ -253,10 +253,10 @@ func TestResolveBurnPreset(t *testing.T) {
 		profile string
 		want    burnPreset
 	}{
-		{profile: "smoke", want: burnPreset{NvidiaDiag: 1, DurationSec: 5 * 60}},
-		{profile: "acceptance", want: burnPreset{NvidiaDiag: 3, DurationSec: 60 * 60}},
-		{profile: "overnight", want: burnPreset{NvidiaDiag: 4, DurationSec: 8 * 60 * 60}},
-		{profile: "", want: burnPreset{NvidiaDiag: 1, DurationSec: 5 * 60}},
+		{profile: "smoke", want: burnPreset{DurationSec: 5 * 60}},
+		{profile: "acceptance", want: burnPreset{DurationSec: 60 * 60}},
+		{profile: "overnight", want: burnPreset{DurationSec: 8 * 60 * 60}},
+		{profile: "", want: burnPreset{DurationSec: 5 * 60}},
 	}
 	for _, tc := range tests {
 		if got := resolveBurnPreset(tc.profile); got != tc.want {

iso/builder/build.sh

@@ -302,6 +302,12 @@ memtest_fail() {
     return 0
 }
 
+nvidia_runtime_fail() {
+    msg="$1"
+    echo "ERROR: ${msg}" >&2
+    exit 1
+}
+
 iso_memtest_present() {
     iso_path="$1"
     iso_files="$(mktemp)"
@@ -439,6 +445,44 @@ validate_iso_memtest() {
     echo "=== memtest validation OK ==="
 }
 
+validate_iso_nvidia_runtime() {
+    iso_path="$1"
+    [ "$BEE_GPU_VENDOR" = "nvidia" ] || return 0
+
+    echo "=== validating NVIDIA runtime in ISO ==="
+
+    [ -f "$iso_path" ] || nvidia_runtime_fail "ISO not found for NVIDIA runtime validation: $iso_path"
+    require_iso_reader "$iso_path" >/dev/null 2>&1 || nvidia_runtime_fail "ISO reader unavailable for NVIDIA runtime validation"
+    command -v unsquashfs >/dev/null 2>&1 || nvidia_runtime_fail "unsquashfs is required for NVIDIA runtime validation"
+
+    squashfs_tmp="$(mktemp)"
+    squashfs_list="$(mktemp)"
+    iso_read_member "$iso_path" live/filesystem.squashfs "$squashfs_tmp" || {
+        rm -f "$squashfs_tmp" "$squashfs_list"
+        nvidia_runtime_fail "failed to extract live/filesystem.squashfs from ISO"
+    }
+    unsquashfs -ll "$squashfs_tmp" > "$squashfs_list" 2>/dev/null || {
+        rm -f "$squashfs_tmp" "$squashfs_list"
+        nvidia_runtime_fail "failed to inspect filesystem.squashfs from ISO"
+    }
+
+    grep -Eq 'usr/bin/dcgmi$' "$squashfs_list" || {
+        rm -f "$squashfs_tmp" "$squashfs_list"
+        nvidia_runtime_fail "dcgmi missing from final NVIDIA ISO"
+    }
+    grep -Eq 'usr/bin/nv-hostengine$' "$squashfs_list" || {
+        rm -f "$squashfs_tmp" "$squashfs_list"
+        nvidia_runtime_fail "nv-hostengine missing from final NVIDIA ISO"
+    }
+    grep -Eq 'usr/bin/dcgmproftester([0-9]+)?$' "$squashfs_list" || {
+        rm -f "$squashfs_tmp" "$squashfs_list"
+        nvidia_runtime_fail "dcgmproftester missing from final NVIDIA ISO"
+    }
+
+    rm -f "$squashfs_tmp" "$squashfs_list"
+    echo "=== NVIDIA runtime validation OK ==="
+}
+
 append_memtest_grub_entry() {
     grub_cfg="$1"
     [ -f "$grub_cfg" ] || return 1
@@ -1144,6 +1188,7 @@ if [ -f "$ISO_RAW" ]; then
         fi
     fi
     validate_iso_memtest "$ISO_RAW"
+    validate_iso_nvidia_runtime "$ISO_RAW"
     cp "$ISO_RAW" "$ISO_OUT"
     echo ""
     echo "=== done (${BEE_GPU_VENDOR}) ==="

iso/builder/config/package-lists/bee-nvidia.list.chroot

@@ -1,6 +1,10 @@
-# NVIDIA DCGM (Data Center GPU Manager) — dcgmi diag for acceptance testing.
-# DCGM 4 is packaged per CUDA major. The image ships NVIDIA driver 590 with CUDA 13 userspace,
-# so install the CUDA 13 build plus proprietary diagnostic components explicitly.
+# NVIDIA DCGM (Data Center GPU Manager).
+# Validate uses dcgmi diagnostics; Burn uses dcgmproftester as the official
+# NVIDIA max-compute recipe. The smoketest/runtime contract treats
+# dcgmproftester as required in the LiveCD.
+# DCGM 4 is packaged per CUDA major. The image ships NVIDIA driver 590 with
+# CUDA 13 userspace, so install the CUDA 13 build plus proprietary components
+# explicitly.
 datacenter-gpu-manager-4-cuda13=1:%%DCGM_VERSION%%
 datacenter-gpu-manager-4-proprietary=1:%%DCGM_VERSION%%
 datacenter-gpu-manager-4-proprietary-cuda13=1:%%DCGM_VERSION%%

iso/builder/smoketest.sh

@@ -52,6 +52,31 @@ else
     fail "nvidia-smi: NOT FOUND"
 fi
 
+if p=$(PATH="/usr/local/bin:$PATH" command -v dcgmi 2>/dev/null); then
+    ok "dcgmi found: $p"
+else
+    fail "dcgmi: NOT FOUND"
+fi
+
+if p=$(PATH="/usr/local/bin:$PATH" command -v nv-hostengine 2>/dev/null); then
+    ok "nv-hostengine found: $p"
+else
+    fail "nv-hostengine: NOT FOUND"
+fi
+
+DCGM_PROFTESTER=""
+for tool in dcgmproftester dcgmproftester13 dcgmproftester12 dcgmproftester11; do
+    if p=$(PATH="/usr/local/bin:$PATH" command -v "$tool" 2>/dev/null); then
+        DCGM_PROFTESTER="$p"
+        break
+    fi
+done
+if [ -n "$DCGM_PROFTESTER" ]; then
+    ok "dcgmproftester found: $DCGM_PROFTESTER"
+else
+    fail "dcgmproftester: NOT FOUND"
+fi
+
 for tool in bee-gpu-burn bee-john-gpu-stress bee-nccl-gpu-stress all_reduce_perf; do
     if p=$(PATH="/usr/local/bin:$PATH" command -v "$tool" 2>/dev/null); then
         ok "$tool found: $p"
@@ -60,6 +85,12 @@ for tool in bee-gpu-burn bee-john-gpu-stress bee-nccl-gpu-stress all_reduce_perf
     fi
 done
 
+if p=$(PATH="/usr/local/bin:$PATH" command -v nvbandwidth 2>/dev/null); then
+    ok "nvbandwidth found: $p"
+else
+    warn "nvbandwidth: NOT FOUND"
+fi
+
 echo ""
 echo "-- NVIDIA modules --"
 KO_DIR="/usr/local/lib/nvidia"