diff --git a/cadvisor.go b/cadvisor.go
index f35af67fff..ff84b4d516 100644
--- a/cadvisor.go
+++ b/cadvisor.go
@@ -72,6 +72,7 @@ var (
 		container.NetworkUsageMetrics:    struct{}{},
 		container.NetworkTcpUsageMetrics: struct{}{},
 		container.NetworkUdpUsageMetrics: struct{}{},
+		container.PerCpuUsageMetrics:     struct{}{},
 	}
 )
 
@@ -103,7 +104,7 @@ func (ml *metricSetValue) Set(value string) error {
 }
 
 func init() {
-	flag.Var(&ignoreMetrics, "disable_metrics", "comma-separated list of `metrics` to be disabled. Options are 'disk', 'network', 'tcp', 'udp'. Note: tcp and udp are disabled by default due to high CPU usage.")
+	flag.Var(&ignoreMetrics, "disable_metrics", "comma-separated list of `metrics` to be disabled. Options are 'disk', 'network', 'tcp', 'udp', 'percpu'. Note: tcp and udp are disabled by default due to high CPU usage.")
 
 	// Default logging verbosity to V(2)
 	flag.Set("v", "2")
diff --git a/container/factory.go b/container/factory.go
index befb4a9e63..07445f9dd7 100644
--- a/container/factory.go
+++ b/container/factory.go
@@ -42,6 +42,7 @@ type MetricKind string
 
 const (
 	CpuUsageMetrics        MetricKind = "cpu"
+	PerCpuUsageMetrics     MetricKind = "percpu"
 	MemoryUsageMetrics     MetricKind = "memory"
 	CpuLoadMetrics         MetricKind = "cpuLoad"
 	DiskIOMetrics          MetricKind = "diskIO"
diff --git a/container/libcontainer/helpers.go b/container/libcontainer/helpers.go
index c2194ac34a..2f9270758b 100644
--- a/container/libcontainer/helpers.go
+++ b/container/libcontainer/helpers.go
@@ -113,7 +113,8 @@ func GetStats(cgroupManager cgroups.Manager, rootFs string, pid int, ignoreMetri
 	libcontainerStats := &libcontainer.Stats{
 		CgroupStats: cgroupStats,
 	}
-	stats := newContainerStats(libcontainerStats)
+	withPerCPU := !ignoreMetrics.Has(container.PerCpuUsageMetrics)
+	stats := newContainerStats(libcontainerStats, withPerCPU)
 
 	// If we know the pid then get network stats from /proc/<pid>/net/dev
 	if pid == 0 {
@@ -467,14 +468,17 @@ func minUint32(x, y uint32) uint32 {
 var numCpusFunc = getNumberOnlineCPUs
 
 // Convert libcontainer stats to info.ContainerStats.
-func setCpuStats(s *cgroups.Stats, ret *info.ContainerStats) {
+func setCpuStats(s *cgroups.Stats, ret *info.ContainerStats, withPerCPU bool) {
 	ret.Cpu.Usage.User = s.CpuStats.CpuUsage.UsageInUsermode
 	ret.Cpu.Usage.System = s.CpuStats.CpuUsage.UsageInKernelmode
-	ret.Cpu.Usage.Total = 0
+	ret.Cpu.Usage.Total = s.CpuStats.CpuUsage.TotalUsage
 	ret.Cpu.CFS.Periods = s.CpuStats.ThrottlingData.Periods
 	ret.Cpu.CFS.ThrottledPeriods = s.CpuStats.ThrottlingData.ThrottledPeriods
 	ret.Cpu.CFS.ThrottledTime = s.CpuStats.ThrottlingData.ThrottledTime
 
+	if !withPerCPU {
+		return
+	}
 	if len(s.CpuStats.CpuUsage.PercpuUsage) == 0 {
 		// libcontainer's 'GetStats' can leave 'PercpuUsage' nil if it skipped the
 		// cpuacct subsystem.
@@ -501,7 +505,6 @@ func setCpuStats(s *cgroups.Stats, ret *info.ContainerStats) {
 
 	for i := uint32(0); i < numActual; i++ {
 		ret.Cpu.Usage.PerCpu[i] = s.CpuStats.CpuUsage.PercpuUsage[i]
-		ret.Cpu.Usage.Total += s.CpuStats.CpuUsage.PercpuUsage[i]
 	}
 
 }
@@ -587,13 +590,13 @@ func setNetworkStats(libcontainerStats *libcontainer.Stats, ret *info.ContainerS
 	}
 }
 
-func newContainerStats(libcontainerStats *libcontainer.Stats) *info.ContainerStats {
+func newContainerStats(libcontainerStats *libcontainer.Stats, withPerCPU bool) *info.ContainerStats {
 	ret := &info.ContainerStats{
 		Timestamp: time.Now(),
 	}
 
 	if s := libcontainerStats.CgroupStats; s != nil {
-		setCpuStats(s, ret)
+		setCpuStats(s, ret, withPerCPU)
 		setDiskIoStats(s, ret)
 		setMemoryStats(s, ret)
 	}
diff --git a/container/libcontainer/helpers_test.go b/container/libcontainer/helpers_test.go
index ffde8e2bd6..b2d15a502f 100644
--- a/container/libcontainer/helpers_test.go
+++ b/container/libcontainer/helpers_test.go
@@ -122,7 +122,7 @@ func TestMorePossibleCPUs(t *testing.T) {
 		},
 	}
 	var ret info.ContainerStats
-	setCpuStats(s, &ret)
+	setCpuStats(s, &ret, true)
 
 	expected := info.ContainerStats{
 		Cpu: info.CpuStats{
@@ -130,7 +130,7 @@ func TestMorePossibleCPUs(t *testing.T) {
 				PerCpu: perCpuUsage[0:realNumCPUs],
 				User:   s.CpuStats.CpuUsage.UsageInUsermode,
 				System: s.CpuStats.CpuUsage.UsageInKernelmode,
-				Total:  8562955455524 * uint64(realNumCPUs),
+				Total:  33802947350272,
 			},
 		},
 	}
diff --git a/metrics/prometheus.go b/metrics/prometheus.go
index 2dd7747b83..4354071b81 100644
--- a/metrics/prometheus.go
+++ b/metrics/prometheus.go
@@ -150,10 +150,18 @@ func NewPrometheusCollector(i infoProvider, f ContainerLabelsFunc) *PrometheusCo
 				},
 			}, {
 				name:        "container_cpu_usage_seconds_total",
-				help:        "Cumulative cpu time consumed per cpu in seconds.",
+				help:        "Cumulative cpu time consumed in seconds.",
 				valueType:   prometheus.CounterValue,
 				extraLabels: []string{"cpu"},
 				getValues: func(s *info.ContainerStats) metricValues {
+					if len(s.Cpu.Usage.PerCpu) == 0 {
+						if s.Cpu.Usage.Total > 0 {
+							return metricValues{{
+								value:  float64(s.Cpu.Usage.Total) / float64(time.Second),
+								labels: []string{"total"},
+							}}
+						}
+					}
 					values := make(metricValues, 0, len(s.Cpu.Usage.PerCpu))
 					for i, value := range s.Cpu.Usage.PerCpu {
 						if value > 0 {
diff --git a/metrics/testdata/prometheus_metrics b/metrics/testdata/prometheus_metrics
index 5b8d0ac9bd..26438abd86 100644
--- a/metrics/testdata/prometheus_metrics
+++ b/metrics/testdata/prometheus_metrics
@@ -28,7 +28,7 @@ container_cpu_load_average_10s{container_env_foo_env="prod",container_label_foo_
 # HELP container_cpu_system_seconds_total Cumulative system cpu time consumed in seconds.
 # TYPE container_cpu_system_seconds_total counter
 container_cpu_system_seconds_total{container_env_foo_env="prod",container_label_foo_label="bar",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 7e-09
-# HELP container_cpu_usage_seconds_total Cumulative cpu time consumed per cpu in seconds.
+# HELP container_cpu_usage_seconds_total Cumulative cpu time consumed in seconds.
 # TYPE container_cpu_usage_seconds_total counter
 container_cpu_usage_seconds_total{container_env_foo_env="prod",container_label_foo_label="bar",cpu="cpu00",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 2e-09
 container_cpu_usage_seconds_total{container_env_foo_env="prod",container_label_foo_label="bar",cpu="cpu01",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 3e-09