Refine burn UI and NVIDIA stress flows

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)
+	runNvidiaFn               func(string) (string, error)
-	runNvidiaBenchmarkFn func(string, platform.NvidiaBenchmarkOptions) (string, error)
+	runNvidiaBenchmarkFn      func(string, platform.NvidiaBenchmarkOptions) (string, error)
-	runNvidiaStressFn    func(string, platform.NvidiaStressOptions) (string, error)
+	runNvidiaStressFn         func(string, platform.NvidiaStressOptions) (string, error)
-	runMemoryFn          func(string) (string, error)
+	runNvidiaComputeFn        func(string, int, []int) (string, error)
-	runStorageFn         func(string) (string, error)
+	runNvidiaPowerFn          func(string, int, []int) (string, error)
-	runCPUFn             func(string, int) (string, error)
+	runNvidiaPulseFn          func(string, int, []int) (string, error)
-	detectVendorFn       func() string
+	runNvidiaBandwidthFn      func(string, []int) (string, error)
-	listAMDGPUsFn        func() ([]platform.AMDGPUInfo, error)
+	runNvidiaTargetedStressFn func(string, int, []int) (string, error)
-	runAMDPackFn         func(string) (string, error)
+	runMemoryFn               func(string) (string, error)
-	listNvidiaGPUsFn     func() ([]platform.NvidiaGPU, 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"`
+	ComponentKey  string                 `json:"component_key"`
-	Status        string                  `json:"status"` // "OK", "Warning", "Critical", "Unknown"
+	Status        string                 `json:"status"` // "OK", "Warning", "Critical", "Unknown"
-	LastCheckedAt time.Time               `json:"last_checked_at"`
+	LastCheckedAt time.Time              `json:"last_checked_at"`
-	LastChangedAt time.Time               `json:"last_changed_at"`
+	LastChangedAt time.Time              `json:"last_changed_at"`
-	ErrorSummary  string                  `json:"error_summary,omitempty"`
+	ErrorSummary  string                 `json:"error_summary,omitempty"`
-	History       []ComponentStatusEntry  `json:"history"`
+	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: "nvidia-compute", Available: nvidiaUp && profErr == nil, Vendor: "nvidia"},
-		{ID: "dcgm", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
+		{ID: "nvidia-targeted-power", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
-		{ID: "john", Available: nvidiaUp, Vendor: "nvidia"},
+		{ID: "nvidia-pulse", Available: nvidiaUp && dcgmErr == nil, Vendor: "nvidia"},
-		{ID: "nccl", Available: nvidiaUp, 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/pages.go

@@ -689,7 +689,7 @@ func renderValidate() string {
 </div>
 
 <div class="grid3">
-` + renderSATCard("nvidia", "NVIDIA GPU", `<div class="form-row"><label>Diag Level</label><select id="sat-nvidia-level"><option value="1">Level 1 — Quick</option><option value="2">Level 2 — Standard</option><option value="3">Level 3 — Extended</option><option value="4">Level 4 — Full</option></select></div>`) +
+` + renderSATCard("nvidia", "NVIDIA GPU", `<div class="form-row"><label>Diag Level</label><select id="sat-nvidia-level"><option value="1">Level 1 — Quick</option><option value="2">Level 2 — Standard</option><option value="3">Level 3 — Extended</option><option value="4">Level 4 — Full</option></select></div><div style="display:flex;gap:8px;flex-wrap:wrap;margin-bottom:8px"><button id="sat-btn-nvidia-targeted-stress" class="btn" type="button" onclick="runSAT('nvidia-targeted-stress')">Targeted Stress (dcgmi diag targeted_stress)</button></div><p style="color:var(--muted);font-size:12px;margin:0">Official DCGM `+"targeted_stress"+` stays in Validate as a controlled diagnostic load, not a max-burn recipe.</p>`) +
 		renderSATCard("memory", "Memory", "") +
 		renderSATCard("storage", "Storage", "") +
 		renderSATCard("cpu", "CPU", `<div class="form-row"><label>Duration (seconds)</label><input type="number" id="sat-cpu-dur" value="60" min="10"></div>`) +
@@ -708,9 +708,10 @@ let satES = null;
 function runSAT(target) {
   if (satES) { satES.close(); satES = null; }
   const body = {};
-  const labels = {nvidia:'Validate GPU', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
+  const labels = {nvidia:'Validate GPU', 'nvidia-targeted-stress':'NVIDIA Targeted Stress Validate (dcgmi diag targeted_stress)', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
   body.display_name = labels[target] || ('Validate ' + target);
   if (target === 'nvidia') body.diag_level = parseInt(document.getElementById('sat-nvidia-level').value)||1;
+  if (target === 'nvidia-targeted-stress') body.duration = 300;
   if (target === 'cpu') body.duration = parseInt(document.getElementById('sat-cpu-dur').value)||60;
   document.getElementById('sat-output').style.display='block';
   document.getElementById('sat-title').textContent = '— ' + target;
@@ -727,7 +728,7 @@ function runSAT(target) {
 }
 function runAllSAT() {
   const cycles = Math.max(1, parseInt(document.getElementById('sat-cycles').value)||1);
-  const targets = ['nvidia','memory','storage','cpu','amd','amd-mem','amd-bandwidth'];
+  const targets = ['nvidia','nvidia-targeted-stress','memory','storage','cpu','amd','amd-mem','amd-bandwidth'];
   const total = targets.length * cycles;
   let enqueued = 0;
   const status = document.getElementById('sat-all-status');
@@ -739,9 +740,10 @@ function runAllSAT() {
     const btn = document.getElementById('sat-btn-' + target);
     if (btn && btn.disabled) { enqueueNext(cycle, idx+1); return; }
     const body = {};
-    const labels = {nvidia:'Validate GPU', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
+    const labels = {nvidia:'Validate GPU', 'nvidia-targeted-stress':'NVIDIA Targeted Stress Validate (dcgmi diag targeted_stress)', memory:'Validate Memory', storage:'Validate Storage', cpu:'Validate CPU', amd:'Validate AMD GPU', 'amd-mem':'AMD GPU MEM Integrity', 'amd-bandwidth':'AMD GPU MEM Bandwidth'};
     body.display_name = labels[target] || ('Validate ' + target);
     if (target === 'nvidia') body.diag_level = parseInt(document.getElementById('sat-nvidia-level').value)||1;
+    if (target === 'nvidia-targeted-stress') body.duration = 300;
     if (target === 'cpu') body.duration = parseInt(document.getElementById('sat-cpu-dur').value)||60;
     fetch('/api/sat/'+target+'/run', {method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify(body)})
       .then(r=>r.json()).then(()=>{
@@ -756,6 +758,7 @@ function runAllSAT() {
 <script>
 fetch('/api/gpu/presence').then(r=>r.json()).then(gp => {
     if (!gp.nvidia) disableSATCard('nvidia', 'No NVIDIA GPU detected');
+    if (!gp.nvidia) disableSATCard('nvidia-targeted-stress', 'No NVIDIA GPU detected');
     if (!gp.amd) disableSATCard('amd', 'No AMD GPU detected');
     if (!gp.amd) disableSATCard('amd-mem', 'No AMD GPU detected');
     if (!gp.amd) disableSATCard('amd-bandwidth', 'No AMD GPU detected');
@@ -976,62 +979,91 @@ benchmarkLoadGPUs();
 // ── Burn ──────────────────────────────────────────────────────────────────────
 
 func renderBurn() string {
-	return `<div class="alert alert-warn" style="margin-bottom:16px"><strong>&#9888; Warning:</strong> Stress tests on this page run hardware at maximum load. Repeated or prolonged use may reduce hardware lifespan (storage endurance, GPU wear). Use only when necessary.</div>
+	return `<div class="alert alert-warn" style="margin-bottom:16px"><strong>&#9888; Warning:</strong> Stress tests on this page run hardware at high load. Repeated or prolonged use may reduce hardware lifespan. Use only when necessary.</div>
+<div class="alert alert-info" style="margin-bottom:16px"><strong>Scope:</strong> DCGM diagnostics and ` + "targeted_stress" + ` remain in <a href="/validate">Validate</a>. Burn exposes official NVIDIA load recipes by test goal plus separate custom stress tools.</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">Burn Profile</div>
-  <div class="card-body" style="display:flex;align-items:center;gap:16px;flex-wrap:wrap">
+  <div class="card-body burn-profile-body">
-    <div class="form-row" style="margin:0;max-width:380px"><label>Preset</label><select id="burn-profile">
+    <div class="burn-profile-col">
-      <option value="smoke" selected>Smoke — quick check (~5 min)</option>
+      <div class="form-row" style="margin:0 0 8px"><label>Preset</label></div>
-      <option value="acceptance">Acceptance — 1 hour</option>
+      <label class="cb-row"><input type="radio" name="burn-profile" value="smoke" checked><span>Smoke — quick check (~5 min)</span></label>
-      <option value="overnight">Overnight — 8 hours</option>
+      <label class="cb-row"><input type="radio" name="burn-profile" value="acceptance"><span>Acceptance — 1 hour</span></label>
-    </select></div>
+      <label class="cb-row"><input type="radio" name="burn-profile" value="overnight"><span>Overnight — 8 hours</span></label>
-    <button class="btn btn-primary" onclick="runAll()">&#9654; Run All</button>
+    </div>
+    <div class="burn-profile-col burn-profile-action">
+      <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>
+      <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>
+  <div class="card-body" style="padding-top:0;display:flex;justify-content:center">
     <span id="burn-all-status" style="font-size:12px;color:var(--muted)"></span>
   </div>
 </div>
 
-<div class="grid3" style="margin-bottom:16px">
+<div class="card" style="margin-bottom:16px">
-
+  <div class="card-head">NVIDIA GPU Selection</div>
-<div class="card">
-  <div class="card-head">GPU Stress</div>
   <div class="card-body">
-    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">NVIDIA tools run on all discovered GPUs. DCGM is the official NVIDIA diagnostic path. NCCL exercises multi-GPU fabric and is not a full compute burn.</p>
+    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Official NVIDIA recipes and custom NVIDIA stressors use only the GPUs selected here. Multi-GPU interconnect tests are limited to this selection as well.</p>
-    <div id="gpu-tools-list">
+    <div style="display:flex;gap:8px;flex-wrap:wrap;margin-bottom:8px">
-      <label class="cb-row"><input type="checkbox" id="burn-gpu-bee" value="bee-gpu-burn" disabled><span>bee-gpu-burn <span class="cb-note" id="note-bee"></span></span></label>
+      <button class="btn btn-sm btn-secondary" type="button" onclick="burnSelectAll()">Select All</button>
-      <label class="cb-row"><input type="checkbox" id="burn-gpu-dcgm" value="dcgm" disabled><span>DCGM Diagnostics (Official NVIDIA) <span class="cb-note" id="note-dcgm"></span></span></label>
+      <button class="btn btn-sm btn-secondary" type="button" onclick="burnSelectNone()">Clear</button>
-      <label class="cb-row"><input type="checkbox" id="burn-gpu-john" value="john" disabled><span>John the Ripper (OpenCL) <span class="cb-note" id="note-john"></span></span></label>
-      <label class="cb-row"><input type="checkbox" id="burn-gpu-nccl" value="nccl" disabled><span>NCCL all_reduce_perf (Interconnect) <span class="cb-note" id="note-nccl"></span></span></label>
-      <label class="cb-row"><input type="checkbox" id="burn-gpu-rvs" value="rvs" disabled><span>RVS GST (AMD) <span class="cb-note" id="note-rvs"></span></span></label>
     </div>
-    <button class="btn btn-primary" style="margin-top:10px" onclick="runGPUStress()">&#9654; Run GPU Stress</button>
+    <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 class="card">
+<div class="burn-section">Core Burn Paths</div>
-  <div class="card-head">Compute Stress</div>
+<div class="grid2 burn-grid" style="margin-bottom:16px">
-  <div class="card-body">
+<div class="card burn-card">
+  <div class="card-head card-head-actions"><span>GPU Max Load</span><button class="btn btn-primary btn-sm" onclick="runBurnTaskSet([{id:'burn-nvidia-compute',target:'nvidia-compute',label:'NVIDIA Max Compute Load (dcgmproftester)',nvidia:true},{id:'burn-gpu-bee',target:'nvidia-stress',label:'GPU Burn (bee-gpu-burn)',nvidia:true,extra:{loader:'builtin'}},{id:'burn-gpu-john',target:'nvidia-stress',label:'John GPU Stress (john/OpenCL)',nvidia:true,extra:{loader:'john'}},{id:'burn-gpu-rvs',target:'amd-stress',label:'AMD GPU Stress (rvs gst)'}])">Run</button></div>
+  <div class="card-body burn-card-body">
+    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Combine vendor-backed and custom GPU max-load recipes in one run set. ` + "dcgmproftester" + ` is the primary official NVIDIA path; custom stressors remain available as parallel checkbox options.</p>
+    <label class="cb-row"><input type="checkbox" id="burn-nvidia-compute" checked disabled><span>NVIDIA Max Compute Load (dcgmproftester) <span class="cb-note" id="note-nvidia-compute"></span></span></label>
+    <label class="cb-row"><input type="checkbox" id="burn-gpu-bee" checked disabled><span>GPU Burn (bee-gpu-burn) <span class="cb-note" id="note-bee"></span></span></label>
+    <label class="cb-row"><input type="checkbox" id="burn-gpu-john" disabled><span>John GPU Stress (john/OpenCL) <span class="cb-note" id="note-john"></span></span></label>
+    <label class="cb-row"><input type="checkbox" id="burn-gpu-rvs" disabled><span>AMD GPU Stress (rvs gst) <span class="cb-note" id="note-rvs"></span></span></label>
+  </div>
+</div>
+
+<div class="card burn-card">
+  <div class="card-head card-head-actions"><span>Compute Stress</span><button class="btn btn-primary btn-sm" onclick="runBurnTaskSet([{id:'burn-cpu',target:'cpu',label:'CPU Burn-in'},{id:'burn-mem-stress',target:'memory-stress',label:'Memory Burn-in'},{id:'burn-sat-stress',target:'sat-stress',label:'SAT Stress (stressapptest)'}])">Run</button></div>
+  <div class="card-body burn-card-body">
     <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Select which subsystems to stress. Each checked item runs as a separate task.</p>
     <label class="cb-row"><input type="checkbox" id="burn-cpu" checked><span>CPU stress (stress-ng)</span></label>
     <label class="cb-row"><input type="checkbox" id="burn-mem-stress" checked><span>Memory stress (stress-ng --vm)</span></label>
     <label class="cb-row"><input type="checkbox" id="burn-sat-stress"><span>stressapptest (CPU + memory bus)</span></label>
-    <button class="btn btn-primary" style="margin-top:10px" onclick="runComputeStress()">&#9654; Run Compute Stress</button>
+  </div>
+</div>
+</div>
+
+<div class="burn-section">GPU-Specific Tests</div>
+<div class="grid2 burn-grid" style="margin-bottom:16px">
+<div class="card burn-card">
+  <div class="card-head card-head-actions"><span>Power Delivery / Power Budget</span><button class="btn btn-primary btn-sm" onclick="runBurnTaskSet([{id:'burn-nvidia-power',target:'nvidia-targeted-power',label:'NVIDIA Targeted Power (dcgmi diag targeted_power)',nvidia:true},{id:'burn-nvidia-pulse',target:'nvidia-pulse',label:'NVIDIA Pulse Test (dcgmi diag pulse_test)',nvidia:true}])">Run</button></div>
+  <div class="card-body burn-card-body">
+    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Official NVIDIA power-oriented recipes. ` + "targeted_power" + ` checks sustained delivery; ` + "pulse_test" + ` checks transient behavior.</p>
+    <label class="cb-row"><input type="checkbox" id="burn-nvidia-power" disabled><span>NVIDIA Targeted Power (dcgmi diag targeted_power) <span class="cb-note" id="note-nvidia-power"></span></span></label>
+    <label class="cb-row"><input type="checkbox" id="burn-nvidia-pulse" disabled><span>NVIDIA Pulse Test (dcgmi diag pulse_test) <span class="cb-note" id="note-nvidia-pulse"></span></span></label>
   </div>
 </div>
 
-<div class="card">
+<div class="card burn-card">
-  <div class="card-head">Platform Thermal Cycling</div>
+  <div class="card-head card-head-actions"><span>Interconnect / Bandwidth</span><button class="btn btn-primary btn-sm" onclick="runBurnTaskSet([{id:'burn-nvidia-interconnect',target:'nvidia-interconnect',label:'NVIDIA Interconnect Test (NCCL all_reduce_perf)',nvidia:true},{id:'burn-nvidia-bandwidth',target:'nvidia-bandwidth',label:'NVIDIA Bandwidth Test (NVBandwidth)',nvidia:true}])">Run</button></div>
-  <div class="card-body">
+  <div class="card-body burn-card-body">
-    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Repeated load+idle cycles. Detects cooling recovery failures and GPU throttle. Smoke: 2×90s. Acceptance: 4×300s.</p>
+    <p style="font-size:12px;color:var(--muted);margin:0 0 10px">Official NVIDIA fabric paths. NCCL is interconnect-only and is not a compute burn. NVBandwidth validates copy and bandwidth paths.</p>
-    <p style="font-size:12px;font-weight:600;margin:0 0 6px">Load components:</p>
+    <label class="cb-row"><input type="checkbox" id="burn-nvidia-interconnect" disabled><span>NVIDIA Interconnect Test (NCCL all_reduce_perf) <span class="cb-note" id="note-nvidia-interconnect"></span></span></label>
-    <label class="cb-row"><input type="checkbox" id="burn-pt-cpu" checked><span>CPU (stressapptest)</span></label>
+    <label class="cb-row"><input type="checkbox" id="burn-nvidia-bandwidth" disabled><span>NVIDIA Bandwidth Test (NVBandwidth) <span class="cb-note" id="note-nvidia-bandwidth"></span></span></label>
-    <label class="cb-row"><input type="checkbox" id="burn-pt-nvidia" disabled><span>NVIDIA GPU <span class="cb-note" id="note-pt-nvidia"></span></span></label>
-    <label class="cb-row"><input type="checkbox" id="burn-pt-amd" disabled><span>AMD GPU <span class="cb-note" id="note-pt-amd"></span></span></label>
-    <button class="btn btn-primary" style="margin-top:10px" onclick="runPlatformStress()">&#9654; Run Thermal Cycling</button>
   </div>
 </div>
-
 </div>
 
 <div id="bi-output" style="display:none;margin-top:16px" class="card">
@@ -1040,154 +1072,224 @@ func renderBurn() string {
 </div>
 
 <style>
-.cb-row { display:flex; align-items:center; gap:8px; padding:4px 0; cursor:pointer; font-size:13px; }
+.cb-row { display:flex; align-items:flex-start; gap:8px; padding:4px 0; cursor:pointer; font-size:13px; }
-.cb-row input[type=checkbox] { width:16px; height:16px; flex-shrink:0; }
+.cb-row input[type=checkbox] { width:16px; height:16px; margin-top:2px; flex-shrink:0; }
 .cb-row input[type=checkbox]:disabled { opacity:0.4; cursor:not-allowed; }
 .cb-row input[type=checkbox]:disabled ~ span { opacity:0.45; cursor:not-allowed; }
 .cb-note { font-size:11px; color:var(--muted); font-style:italic; }
+.burn-gpu-row { display:flex; align-items:flex-start; gap:8px; padding:6px 0; cursor:pointer; font-size:13px; }
+.burn-gpu-row input[type=checkbox] { width:16px; height:16px; margin-top:2px; flex-shrink:0; }
+.burn-profile-body { display:grid; grid-template-columns:1fr 1fr 1fr; gap:24px; align-items:stretch; }
+.burn-profile-col { min-width:0; }
+.burn-profile-action { display:flex; flex-direction:column; align-items:center; justify-content:flex-start; gap:8px; }
+.burn-profile-action p { font-size:12px; color:var(--muted); margin:0; width:100%; text-align:left; }
+.burn-section { font-size:12px; font-weight:700; letter-spacing:.06em; text-transform:uppercase; color:var(--muted); margin:0 0 10px; padding-top:4px; }
+.burn-grid { align-items:stretch; }
+.burn-card { height:100%; display:flex; flex-direction:column; }
+.burn-card-body { flex:1; display:flex; flex-direction:column; }
+.card-head-actions { justify-content:space-between; }
+.card-head-buttons { display:flex; align-items:center; gap:8px; margin-left:auto; }
+@media(max-width:900px){ .card-head-actions { align-items:flex-start; flex-direction:column; } .card-head-buttons { margin-left:0; } .burn-profile-body { grid-template-columns:1fr; } }
 </style>
 
 <script>
 let biES = null;
 
-function profile() { return document.getElementById('burn-profile').value || 'smoke'; }
+function burnProfile() {
+  const selected = document.querySelector('input[name="burn-profile"]:checked');
+  return selected ? selected.value : 'smoke';
+}
 
-function enqueueTask(target, extra) {
+function burnSelectedGPUIndices() {
-  const body = Object.assign({ profile: profile() }, extra || {});
+  return Array.from(document.querySelectorAll('.burn-gpu-checkbox'))
-  return fetch('/api/sat/'+target+'/run', {
+    .filter(function(el) { return el.checked && !el.disabled; })
-    method: 'POST', headers: {'Content-Type':'application/json'}, body: JSON.stringify(body)
+    .map(function(el) { return parseInt(el.value, 10); })
-  }).then(r => r.json());
+    .filter(function(v) { return !Number.isNaN(v); })
+    .sort(function(a, b) { return a - b; });
+}
+
+function burnUpdateSelectionNote() {
+  const note = document.getElementById('burn-selection-note');
+  const selected = burnSelectedGPUIndices();
+  if (!selected.length) {
+    note.textContent = 'Select at least one NVIDIA GPU to enable NVIDIA burn recipes.';
+    return;
+  }
+  note.textContent = 'Selected NVIDIA GPUs: ' + selected.join(', ') + '. Official and custom NVIDIA tasks will use only these GPUs.';
+}
+
+function burnRenderGPUList(gpus) {
+  const root = document.getElementById('burn-gpu-list');
+  if (!gpus || !gpus.length) {
+    root.innerHTML = '<p style="color:var(--muted);font-size:13px">No NVIDIA GPUs detected.</p>';
+    burnUpdateSelectionNote();
+    return;
+  }
+  root.innerHTML = gpus.map(function(gpu) {
+    const mem = gpu.memory_mb > 0 ? ' · ' + gpu.memory_mb + ' MiB' : '';
+    return '<label class="burn-gpu-row">'
+      + '<input class="burn-gpu-checkbox" type="checkbox" value="' + gpu.index + '" checked onchange="burnUpdateSelectionNote()">'
+      + '<span><strong>GPU ' + gpu.index + '</strong> — ' + gpu.name + mem + '</span>'
+      + '</label>';
+  }).join('');
+  burnUpdateSelectionNote();
+}
+
+function burnSelectAll() {
+  document.querySelectorAll('.burn-gpu-checkbox').forEach(function(el) { el.checked = true; });
+  burnUpdateSelectionNote();
+}
+
+function burnSelectNone() {
+  document.querySelectorAll('.burn-gpu-checkbox').forEach(function(el) { el.checked = false; });
+  burnUpdateSelectionNote();
+}
+
+function burnLoadGPUs() {
+  fetch('/api/gpu/nvidia').then(function(r) {
+    return r.json().then(function(body) {
+      if (!r.ok) throw new Error(body.error || ('HTTP ' + r.status));
+      return body;
+    });
+  }).then(function(gpus) {
+    burnRenderGPUList(gpus);
+  }).catch(function(err) {
+    document.getElementById('burn-gpu-list').innerHTML = '<p style="color:var(--crit-fg);font-size:13px">Error: ' + err.message + '</p>';
+    burnUpdateSelectionNote();
+  });
+}
+
+function enqueueBurnTask(target, label, extra, useSelectedNvidia) {
+  const body = Object.assign({ profile: burnProfile(), display_name: label }, extra || {});
+  if (useSelectedNvidia) {
+    const selected = burnSelectedGPUIndices();
+    if (!selected.length) {
+      return Promise.reject(new Error('Select at least one NVIDIA GPU.'));
+    }
+    body.gpu_indices = selected;
+  }
+  return fetch('/api/sat/' + target + '/run', {
+    method: 'POST',
+    headers: {'Content-Type':'application/json'},
+    body: JSON.stringify(body)
+  }).then(function(r) {
+    return r.json().then(function(payload) {
+      if (!r.ok) throw new Error(payload.error || ('HTTP ' + r.status));
+      return payload;
+    });
+  });
 }
 
 function streamTask(taskId, label) {
   if (biES) { biES.close(); biES = null; }
   document.getElementById('bi-output').style.display = 'block';
-  document.getElementById('bi-title').textContent = '— ' + label + ' [' + profile() + ']';
+  document.getElementById('bi-title').textContent = '— ' + label + ' [' + burnProfile() + ']';
   const term = document.getElementById('bi-terminal');
   term.textContent = 'Task ' + taskId + ' queued. Streaming...\n';
-  biES = new EventSource('/api/tasks/'+taskId+'/stream');
+  biES = new EventSource('/api/tasks/' + taskId + '/stream');
-  biES.onmessage = e => { term.textContent += e.data+'\n'; term.scrollTop = term.scrollHeight; };
+  biES.onmessage = function(e) { term.textContent += e.data + '\n'; term.scrollTop = term.scrollHeight; };
-  biES.addEventListener('done', e => {
+  biES.addEventListener('done', function(e) {
-    biES.close(); biES = null;
+    biES.close();
-    term.textContent += (e.data ? '\nERROR: '+e.data : '\nCompleted.')+'\n';
+    biES = null;
+    term.textContent += (e.data ? '\nERROR: ' + e.data : '\nCompleted.') + '\n';
+    term.scrollTop = term.scrollHeight;
   });
 }
 
-function runGPUStress() {
+function runBurnTaskSet(tasks, statusElId) {
-  const tasks = [
+  const enabled = tasks.filter(function(t) {
-    {id:'burn-gpu-bee', target:'nvidia-stress', label:'bee-gpu-burn', extra:{loader:'builtin'}},
-    {id:'burn-gpu-dcgm', target:'nvidia', label:'DCGM Diagnostics (Official NVIDIA)', extra:{display_name:'NVIDIA DCGM Diagnostics (Official)'}},
-    {id:'burn-gpu-john', target:'nvidia-stress', label:'John GPU Stress', extra:{loader:'john'}},
-    {id:'burn-gpu-nccl', target:'nvidia-stress', label:'NCCL Interconnect Stress', extra:{loader:'nccl', display_name:'NCCL Interconnect Stress'}},
-    {id:'burn-gpu-rvs', target:'amd-stress', label:'RVS GST', extra:{}},
-  ];
-  tasks.filter(t => {
     const el = document.getElementById(t.id);
     return el && el.checked && !el.disabled;
-  }).forEach(t => {
-    enqueueTask(t.target, t.extra).then(d => { streamTask(d.task_id, t.label); });
   });
-}
+  const status = statusElId ? document.getElementById(statusElId) : null;
-
+  if (status) status.textContent = '';
-function runComputeStress() {
+  if (!enabled.length) {
-  const tasks = [
+    if (status) status.textContent = 'No tasks selected.';
-    {id:'burn-cpu', target:'cpu'},
+    return;
-    {id:'burn-mem-stress', target:'memory-stress'},
+  }
-    {id:'burn-sat-stress', target:'sat-stress'},
+  enabled.forEach(function(t) {
-  ];
+    enqueueBurnTask(t.target, t.label, t.extra, !!t.nvidia)
-  let last = null;
+      .then(function(d) {
-  tasks.filter(t => {
+        if (status) status.textContent = enabled.length + ' task(s) queued.';
-    const el = document.getElementById(t.id);
+        streamTask(d.task_id, t.label);
-    return el && el.checked;
+      })
-  }).forEach(t => {
+      .catch(function(err) {
-    enqueueTask(t.target).then(d => { last = d; streamTask(d.task_id, t.target); });
+        if (status) status.textContent = 'Error: ' + err.message;
+        const term = document.getElementById('bi-terminal');
+        document.getElementById('bi-output').style.display = 'block';
+        term.textContent += 'ERROR: ' + err.message + '\n';
+      });
   });
 }
 
 function runPlatformStress() {
   const comps = [];
-  if (document.getElementById('burn-pt-cpu').checked) comps.push('cpu');
+  const computeIDs = ['burn-cpu', 'burn-mem-stress', 'burn-sat-stress'];
-  const nv = document.getElementById('burn-pt-nvidia');
+  const gpuIDs = ['burn-nvidia-compute', 'burn-gpu-bee', 'burn-gpu-john', 'burn-gpu-rvs'];
-  if (nv && nv.checked && !nv.disabled) comps.push('gpu');
+  const hasChecked = function(ids) {
-  const am = document.getElementById('burn-pt-amd');
+    return ids.some(function(id) {
-  if (am && am.checked && !am.disabled) comps.push('gpu');
+      const el = document.getElementById(id);
+      return el && el.checked && !el.disabled;
+    });
+  };
+  if (hasChecked(computeIDs)) comps.push('cpu');
+  if (hasChecked(gpuIDs)) comps.push('gpu');
+  if (!comps.length) {
+    const status = document.getElementById('burn-all-status');
+    if (status) status.textContent = 'Select at least one test in GPU Max Load or Compute Stress.';
+    return;
+  }
   const extra = comps.length > 0 ? {platform_components: comps} : {};
-  enqueueTask('platform-stress', extra).then(d => streamTask(d.task_id, 'platform-stress'));
+  enqueueBurnTask('platform-stress', 'Platform Thermal Cycling', extra, false).then(function(d) {
-}
+    streamTask(d.task_id, 'Platform Thermal Cycling');
-
-function runAll() {
-  const status = document.getElementById('burn-all-status');
-  status.textContent = 'Enqueuing...';
-  let count = 0;
-  const done = () => { count++; status.textContent = count + ' tasks queued.'; };
-
-  // GPU tests
-  const gpuTasks = [
-    {id:'burn-gpu-bee', target:'nvidia-stress', label:'bee-gpu-burn', extra:{loader:'builtin'}},
-    {id:'burn-gpu-dcgm', target:'nvidia', label:'DCGM Diagnostics (Official NVIDIA)', extra:{display_name:'NVIDIA DCGM Diagnostics (Official)'}},
-    {id:'burn-gpu-john', target:'nvidia-stress', label:'John GPU Stress', extra:{loader:'john'}},
-    {id:'burn-gpu-nccl', target:'nvidia-stress', label:'NCCL Interconnect Stress', extra:{loader:'nccl', display_name:'NCCL Interconnect Stress'}},
-    {id:'burn-gpu-rvs', target:'amd-stress', label:'RVS GST', extra:{}},
-  ];
-  gpuTasks.filter(t => { const el = document.getElementById(t.id); return el && el.checked && !el.disabled; }).forEach(t => {
-    enqueueTask(t.target, t.extra).then(d => { streamTask(d.task_id, t.label); done(); });
   });
-
-  // Compute tests
-  [{id:'burn-cpu',target:'cpu'},{id:'burn-mem-stress',target:'memory-stress'},{id:'burn-sat-stress',target:'sat-stress'}]
-    .filter(t => { const el = document.getElementById(t.id); return el && el.checked; })
-    .forEach(t => enqueueTask(t.target).then(d => { streamTask(d.task_id, t.target); done(); }));
-
-  // Platform
-  const comps = [];
-  if (document.getElementById('burn-pt-cpu').checked) comps.push('cpu');
-  const nv = document.getElementById('burn-pt-nvidia');
-  if (nv && nv.checked && !nv.disabled) comps.push('gpu');
-  const am = document.getElementById('burn-pt-amd');
-  if (am && am.checked && !am.disabled) comps.push('gpu');
-  const ptExtra = comps.length > 0 ? {platform_components: comps} : {};
-  enqueueTask('platform-stress', ptExtra).then(d => { streamTask(d.task_id, 'platform-stress'); done(); });
 }
 
-// Load GPU tool availability
+function runAllBurnTasks() {
-fetch('/api/gpu/tools').then(r => r.json()).then(tools => {
+  const status = document.getElementById('burn-all-status');
-  const nvidiaMap = {'bee-gpu-burn':'burn-gpu-bee','dcgm':'burn-gpu-dcgm','john':'burn-gpu-john','nccl':'burn-gpu-nccl','rvs':'burn-gpu-rvs'};
+  const all = [
-  const noteMap = {'bee-gpu-burn':'note-bee','dcgm':'note-dcgm','john':'note-john','nccl':'note-nccl','rvs':'note-rvs'};
+    {id:'burn-nvidia-compute',target:'nvidia-compute',label:'NVIDIA Max Compute Load (dcgmproftester)',nvidia:true},
-  tools.forEach(t => {
+    {id:'burn-nvidia-power',target:'nvidia-targeted-power',label:'NVIDIA Targeted Power (dcgmi diag targeted_power)',nvidia:true},
-    const cb = document.getElementById(nvidiaMap[t.id]);
+    {id:'burn-nvidia-pulse',target:'nvidia-pulse',label:'NVIDIA Pulse Test (dcgmi diag pulse_test)',nvidia:true},
-    const note = document.getElementById(noteMap[t.id]);
+    {id:'burn-nvidia-interconnect',target:'nvidia-interconnect',label:'NVIDIA Interconnect Test (NCCL all_reduce_perf)',nvidia:true},
+    {id:'burn-nvidia-bandwidth',target:'nvidia-bandwidth',label:'NVIDIA Bandwidth Test (NVBandwidth)',nvidia:true},
+    {id:'burn-gpu-bee',target:'nvidia-stress',label:'GPU Burn (bee-gpu-burn)',nvidia:true,extra:{loader:'builtin'}},
+    {id:'burn-gpu-john',target:'nvidia-stress',label:'John GPU Stress (john/OpenCL)',nvidia:true,extra:{loader:'john'}},
+    {id:'burn-gpu-rvs',target:'amd-stress',label:'AMD GPU Stress (rvs gst)'},
+    {id:'burn-cpu',target:'cpu',label:'CPU Burn-in'},
+    {id:'burn-mem-stress',target:'memory-stress',label:'Memory Burn-in'},
+    {id:'burn-sat-stress',target:'sat-stress',label:'SAT Stress (stressapptest)'},
+  ];
+  status.textContent = 'Enqueuing...';
+  runBurnTaskSet(all, 'burn-all-status');
+}
+
+fetch('/api/gpu/tools').then(function(r) { return r.json(); }).then(function(tools) {
+  const map = {
+    'nvidia-compute': {cb:'burn-nvidia-compute', note:'note-nvidia-compute', reason:'dcgmproftester not available or NVIDIA driver not running'},
+    'nvidia-targeted-power': {cb:'burn-nvidia-power', note:'note-nvidia-power', reason:'dcgmi not available or NVIDIA driver not running'},
+    'nvidia-pulse': {cb:'burn-nvidia-pulse', note:'note-nvidia-pulse', reason:'dcgmi not available or NVIDIA driver not running'},
+    'nvidia-interconnect': {cb:'burn-nvidia-interconnect', note:'note-nvidia-interconnect', reason:'NCCL interconnect tool not available or NVIDIA driver not running'},
+    'nvidia-bandwidth': {cb:'burn-nvidia-bandwidth', note:'note-nvidia-bandwidth', reason:'nvbandwidth or dcgmi not available or NVIDIA driver not running'},
+    'bee-gpu-burn': {cb:'burn-gpu-bee', note:'note-bee', reason:'bee-gpu-burn not available or NVIDIA driver not running'},
+    'john': {cb:'burn-gpu-john', note:'note-john', reason:'bee-john-gpu-stress not available or NVIDIA driver not running'},
+    'rvs': {cb:'burn-gpu-rvs', note:'note-rvs', reason:'AMD driver not running'},
+  };
+  tools.forEach(function(t) {
+    const spec = map[t.id];
+    if (!spec) return;
+    const cb = document.getElementById(spec.cb);
+    const note = document.getElementById(spec.note);
     if (!cb) return;
     if (t.available) {
       cb.disabled = false;
-      if (t.id === 'bee-gpu-burn' || t.id === 'dcgm') cb.checked = true;
+    } else if (note) {
-    } else {
+      note.textContent = '— ' + spec.reason;
-      let reason = t.vendor === 'nvidia' ? 'NVIDIA driver not running' : 'AMD driver not running';
-      if (t.id === 'dcgm' && t.vendor === 'nvidia') reason = 'dcgmi not available or NVIDIA driver not running';
-      if (t.id === 'nccl' && t.vendor === 'nvidia') reason = 'NCCL interconnect tool unavailable or NVIDIA driver not running';
-      if (note) note.textContent = '— ' + reason;
     }
   });
-}).catch(() => {});
+}).catch(function() {});
 
-// Load GPU presence for platform thermal cycling
+burnLoadGPUs();
-fetch('/api/gpu/presence').then(r => r.json()).then(gp => {
-  const nvCb = document.getElementById('burn-pt-nvidia');
-  const amCb = document.getElementById('burn-pt-amd');
-  const nvNote = document.getElementById('note-pt-nvidia');
-  const amNote = document.getElementById('note-pt-amd');
-  if (gp.nvidia) {
-    nvCb.disabled = false;
-    nvCb.checked = true;
-  } else {
-    if (nvNote) nvNote.textContent = '— NVIDIA driver not running';
-  }
-  if (gp.amd) {
-    amCb.disabled = false;
-    amCb.checked = true;
-  } else {
-    if (amNote) amNote.textContent = '— AMD driver not running';
-  }
-}).catch(() => {});
 </script>`
 }
 

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

@@ -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 +650,26 @@ func TestBenchmarkPageRendersGPUSelectionControls(t *testing.T) {
 	}
 }
 
-func TestBurnPageRendersOfficialNVIDIADCGMAndNCCLInterconnectLabel(t *testing.T) {
+func TestValidatePageRendersNvidiaTargetedStressAction(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{
+		`Targeted Stress`,
+		`nvidia-targeted-stress`,
+		`Official DCGM`,
+	} {
+		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 +678,11 @@ func TestBurnPageRendersOfficialNVIDIADCGMAndNCCLInterconnectLabel(t *testing.T)
 	}
 	body := rec.Body.String()
 	for _, needle := range []string{
-		`DCGM Diagnostics (Official NVIDIA)`,
+		`NVIDIA Max Compute Load`,
-		`NCCL all_reduce_perf (Interconnect)`,
+		`dcgmproftester`,
-		`DCGM is the official NVIDIA diagnostic path`,
+		`targeted_stress remain in <a href="/validate">Validate</a>`,
-		`burn-gpu-dcgm`,
+		`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)

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":                 "NVIDIA SAT",
-	"nvidia-benchmark": "NVIDIA Benchmark",
+	"nvidia-targeted-stress": "NVIDIA Targeted Stress Validate (dcgmi diag targeted_stress)",
-	"nvidia-stress":    "NVIDIA GPU Stress",
+	"nvidia-benchmark":       "NVIDIA Benchmark",
-	"memory":           "Memory SAT",
+	"nvidia-compute":         "NVIDIA Max Compute Load (dcgmproftester)",
-	"storage":          "Storage SAT",
+	"nvidia-targeted-power":  "NVIDIA Targeted Power (dcgmi diag targeted_power)",
-	"cpu":              "CPU SAT",
+	"nvidia-pulse":           "NVIDIA Pulse Test (dcgmi diag pulse_test)",
-	"amd":              "AMD GPU SAT",
+	"nvidia-interconnect":    "NVIDIA Interconnect Test (NCCL all_reduce_perf)",
-	"amd-mem":          "AMD GPU MEM Integrity",
+	"nvidia-bandwidth":       "NVIDIA Bandwidth Test (NVBandwidth)",
-	"amd-bandwidth":    "AMD GPU MEM Bandwidth",
+	"nvidia-stress":          "NVIDIA GPU Stress",
-	"amd-stress":       "AMD GPU Burn-in",
+	"memory":                 "Memory SAT",
-	"memory-stress":    "Memory Burn-in",
+	"storage":                "Storage SAT",
-	"sat-stress":       "SAT Stress (stressapptest)",
+	"cpu":                    "CPU SAT",
-	"platform-stress":  "Platform Thermal Cycling",
+	"amd":                    "AMD GPU SAT",
-	"audit":            "Audit",
+	"amd-mem":                "AMD GPU MEM Integrity",
-	"support-bundle":   "Support Bundle",
+	"amd-bandwidth":          "AMD GPU MEM Bandwidth",
-	"install":          "Install to Disk",
+	"amd-stress":             "AMD GPU Burn-in",
-	"install-to-ram":   "Install to RAM",
+	"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{
+		cycles := make([]platform.PlatformStressCycle, 0, len(acceptanceCycles)*8)
-			{LoadSec: 600, IdleSec: 120},
+		for range 8 {
-			{LoadSec: 600, IdleSec: 60},
+			cycles = append(cycles, acceptanceCycles...)
-			{LoadSec: 600, IdleSec: 30},
+		}
-			{LoadSec: 600, IdleSec: 120},
+		return platform.PlatformStressOptions{Cycles: cycles}
-			{LoadSec: 600, IdleSec: 60},
-			{LoadSec: 600, IdleSec: 30},
-			{LoadSec: 600, IdleSec: 120},
-			{LoadSec: 600, IdleSec: 60},
-		}}
 	case "acceptance":
-		return platform.PlatformStressOptions{Cycles: []platform.PlatformStressCycle{
+		return platform.PlatformStressOptions{Cycles: acceptanceCycles}
-			{LoadSec: 300, IdleSec: 60},
-			{LoadSec: 300, IdleSec: 30},
-			{LoadSec: 300, IdleSec: 60},
-			{LoadSec: 300, IdleSec: 30},
-		}}
 	default: // smoke
 		return platform.PlatformStressOptions{Cycles: []platform.PlatformStressCycle{
-			{LoadSec: 90, IdleSec: 60},
+			{LoadSec: 85, IdleSec: 5},
-			{LoadSec: 90, IdleSec: 30},
+			{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: "smoke", want: burnPreset{DurationSec: 5 * 60}},
-		{profile: "acceptance", want: burnPreset{NvidiaDiag: 3, DurationSec: 60 * 60}},
+		{profile: "acceptance", want: burnPreset{DurationSec: 60 * 60}},
-		{profile: "overnight", want: burnPreset{NvidiaDiag: 4, DurationSec: 8 * 60 * 60}},
+		{profile: "overnight", want: burnPreset{DurationSec: 8 * 60 * 60}},
-		{profile: "", want: burnPreset{NvidiaDiag: 1, DurationSec: 5 * 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.
+# NVIDIA DCGM (Data Center GPU Manager).
-# DCGM 4 is packaged per CUDA major. The image ships NVIDIA driver 590 with CUDA 13 userspace,
+# Validate uses dcgmi diagnostics; Burn uses dcgmproftester as the official
-# so install the CUDA 13 build plus proprietary diagnostic components explicitly.
+# 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"