feat: v3.10 GPU stress and NCCL burn updates

audit/internal/platform/nvidia_stress.go

@@ -61,6 +61,20 @@ func buildNvidiaStressJob(opts NvidiaStressOptions) (satJob, error) {
 			collectGPU: true,
 			gpuIndices: selected,
 		}, nil
+	case NvidiaStressLoaderNCCL:
+		cmd := []string{
+			"bee-nccl-gpu-stress",
+			"--seconds", strconv.Itoa(opts.DurationSec),
+		}
+		if len(selected) > 0 {
+			cmd = append(cmd, "--devices", joinIndexList(selected))
+		}
+		return satJob{
+			name:       "03-bee-nccl-gpu-stress.log",
+			cmd:        cmd,
+			collectGPU: true,
+			gpuIndices: selected,
+		}, nil
 	default:
 		return satJob{}, fmt.Errorf("unknown NVIDIA stress loader %q", opts.Loader)
 	}
@@ -78,6 +92,8 @@ func normalizeNvidiaStressOptions(opts *NvidiaStressOptions) {
 		opts.Loader = NvidiaStressLoaderBuiltin
 	case NvidiaStressLoaderJohn:
 		opts.Loader = NvidiaStressLoaderJohn
+	case NvidiaStressLoaderNCCL:
+		opts.Loader = NvidiaStressLoaderNCCL
 	default:
 		opts.Loader = NvidiaStressLoaderBuiltin
 	}

audit/internal/platform/runtime.go

@@ -138,6 +138,8 @@ func (s *System) runtimeToolStatuses(vendor string) []ToolStatus {
 			"nvidia-bug-report.sh",
 			"bee-gpu-burn",
 			"bee-john-gpu-stress",
+			"bee-nccl-gpu-stress",
+			"all_reduce_perf",
 		})...)
 	case "amd":
 		tool := ToolStatus{Name: "rocm-smi"}

audit/internal/platform/sat_test.go

@@ -128,6 +128,40 @@ func TestBuildNvidiaStressJobUsesSelectedLoaderAndDevices(t *testing.T) {
 	}
 }
 
+func TestBuildNvidiaStressJobUsesNCCLLoader(t *testing.T) {
+	t.Parallel()
+
+	oldExecCommand := satExecCommand
+	satExecCommand = func(name string, args ...string) *exec.Cmd {
+		if name == "nvidia-smi" {
+			return exec.Command("sh", "-c", "printf '0\n1\n2\n'")
+		}
+		return exec.Command(name, args...)
+	}
+	t.Cleanup(func() { satExecCommand = oldExecCommand })
+
+	job, err := buildNvidiaStressJob(NvidiaStressOptions{
+		DurationSec: 120,
+		Loader:      NvidiaStressLoaderNCCL,
+		GPUIndices:  []int{2, 0},
+	})
+	if err != nil {
+		t.Fatalf("buildNvidiaStressJob error: %v", err)
+	}
+	wantCmd := []string{"bee-nccl-gpu-stress", "--seconds", "120", "--devices", "0,2"}
+	if len(job.cmd) != len(wantCmd) {
+		t.Fatalf("cmd len=%d want %d (%v)", len(job.cmd), len(wantCmd), job.cmd)
+	}
+	for i := range wantCmd {
+		if job.cmd[i] != wantCmd[i] {
+			t.Fatalf("cmd[%d]=%q want %q", i, job.cmd[i], wantCmd[i])
+		}
+	}
+	if got := joinIndexList(job.gpuIndices); got != "0,2" {
+		t.Fatalf("gpuIndices=%q want 0,2", got)
+	}
+}
+
 func TestEnvIntFallback(t *testing.T) {
 	os.Unsetenv("BEE_MEMTESTER_SIZE_MB")
 	if got := envInt("BEE_MEMTESTER_SIZE_MB", 123); got != 123 {

audit/internal/platform/types.go

@@ -54,6 +54,7 @@ type ToolStatus struct {
 const (
 	NvidiaStressLoaderBuiltin = "builtin"
 	NvidiaStressLoaderJohn    = "john"
+	NvidiaStressLoaderNCCL    = "nccl"
 )
 
 type NvidiaStressOptions struct {