diff --git a/audit/internal/webui/api.go b/audit/internal/webui/api.go
index e756a46..75bf3ca 100644
--- a/audit/internal/webui/api.go
+++ b/audit/internal/webui/api.go
@@ -628,8 +628,10 @@ func (h *handler) handleAPIBenchmarkNvidiaRunKind(target string) http.HandlerFun
 		}
 
 		if rampUp && len(body.GPUIndices) > 1 {
-			// Ramp-up mode: resolve GPU list, then create one task per prefix
-			// [gpu0], [gpu0,gpu1], ..., [gpu0,...,gpuN-1], each running in parallel.
+			// Ramp-up mode: RunNvidiaPowerBench internally ramps from 1 to N GPUs
+			// in Phase 2 (one additional GPU per step). A single task with all
+			// selected GPUs is sufficient — spawning N tasks with growing subsets
+			// would repeat all earlier steps redundantly.
 			gpus, err := apiListNvidiaGPUs(h.opts.App)
 			if err != nil {
 				writeError(w, http.StatusBadRequest, err.Error())
@@ -646,35 +648,27 @@ func (h *handler) handleAPIBenchmarkNvidiaRunKind(target string) http.HandlerFun
 			} else {
 				now := time.Now()
 				rampRunID := fmt.Sprintf("ramp-%s", now.UTC().Format("20060102-150405"))
-				var allTasks []*Task
-				for step := 1; step <= len(resolved); step++ {
-					subset := resolved[:step]
-					stepName := fmt.Sprintf("%s · ramp %d/%d · GPU %s", name, step, len(resolved), formatGPUIndexList(subset))
-					t := &Task{
-						ID:        newJobID("bee-bench-nvidia"),
-						Name:      stepName,
-						Target:    target,
-						Priority:  defaultTaskPriority(target, taskParams{}),
-						Status:    TaskPending,
-						CreatedAt: now,
-						params: taskParams{
-							GPUIndices:       append([]int(nil), subset...),
-							SizeMB:           body.SizeMB,
-							BenchmarkProfile: body.Profile,
-							RunNCCL:          runNCCL && step == len(resolved),
-							ParallelGPUs:     true,
-							RampStep:         step,
-							RampTotal:        len(resolved),
-							RampRunID:        rampRunID,
-							DisplayName:      stepName,
-						},
-					}
-					allTasks = append(allTasks, t)
+				taskName := fmt.Sprintf("%s · ramp 1–%d · GPU %s", name, len(resolved), formatGPUIndexList(resolved))
+				t := &Task{
+					ID:        newJobID("bee-bench-nvidia"),
+					Name:      taskName,
+					Target:    target,
+					Priority:  defaultTaskPriority(target, taskParams{}),
+					Status:    TaskPending,
+					CreatedAt: now,
+					params: taskParams{
+						GPUIndices:       append([]int(nil), resolved...),
+						SizeMB:           body.SizeMB,
+						BenchmarkProfile: body.Profile,
+						RunNCCL:          runNCCL,
+						ParallelGPUs:     true,
+						RampTotal:        len(resolved),
+						RampRunID:        rampRunID,
+						DisplayName:      taskName,
+					},
 				}
-				for _, t := range allTasks {
-					globalQueue.enqueue(t)
-				}
-				writeTaskRunResponse(w, allTasks)
+				globalQueue.enqueue(t)
+				writeTaskRunResponse(w, []*Task{t})
 				return
 			}
 		}