diff --git a/tools/net_speed_benchmark.cpp b/tools/net_speed_benchmark.cpp
index 83fba147..a0e589e5 100644
--- a/tools/net_speed_benchmark.cpp
+++ b/tools/net_speed_benchmark.cpp
@@ -1,31 +1,29 @@
 // Copyright 2013 Yangqing Jia
 
-#include <ctime>
-#include <string>
-#include <vector>
+#include <cuda_runtime.h>
+#include <fcntl.h>
+#include <google/protobuf/text_format.h>
 
-#include "cuda_runtime.h"
-#include "fcntl.h"
-#include "google/protobuf/text_format.h"
+#include <cstring>
+#include <ctime>
 
 #include "caffe/blob.hpp"
 #include "caffe/common.hpp"
 #include "caffe/net.hpp"
 #include "caffe/filler.hpp"
 #include "caffe/proto/caffe.pb.h"
+#include "caffe/util/benchmark.hpp"
 #include "caffe/util/io.hpp"
 #include "caffe/solver.hpp"
 
-using boost::shared_ptr;
-
-using namespace caffe;  // NOLINT(build/namespaces)
+using namespace caffe;
 
 int main(int argc, char** argv) {
+
   int total_iter = 50;
 
   if (argc < 2) {
-    LOG(ERROR) << "net_speed_benchmark net_proto [iterations=50] [CPU/GPU] "
-        << "[Device_id=0]";
+    LOG(ERROR) << "net_speed_benchmark net_proto [iterations=50] [CPU/GPU] [Device_id=0]";
     return 0;
   }
 
@@ -67,52 +65,38 @@ int main(int argc, char** argv) {
   vector<vector<Blob<float>*> >& bottom_vecs = caffe_net.bottom_vecs();
   vector<vector<Blob<float>*> >& top_vecs = caffe_net.top_vecs();
   LOG(ERROR) << "*** Benchmark begins ***";
-  if (Caffe::mode() == Caffe::GPU) {
-    cudaDeviceSynchronize();
-  }
-  clock_t forward_start = clock();
+  Timer total_timer;
+  total_timer.Start();
+  Timer forward_timer;
+  forward_timer.Start();
+  Timer timer;
   for (int i = 0; i < layers.size(); ++i) {
     const string& layername = layers[i]->layer_param().name();
-    if (Caffe::mode() == Caffe::GPU) {
-      cudaDeviceSynchronize();
-    }
-    clock_t start = clock();
+    timer.Start();
     for (int j = 0; j < total_iter; ++j) {
       layers[i]->Forward(bottom_vecs[i], &top_vecs[i]);
     }
-    if (Caffe::mode() == Caffe::GPU) {
-      cudaDeviceSynchronize();
-    }
-    LOG(ERROR) << layername << "\tforward: "
-        << static_cast<float>(clock() - start) / CLOCKS_PER_SEC
-        << " seconds.";
+    timer.Stop();
+    LOG(ERROR) << layername << "\tforward: " << timer.ElapsedSeconds() << " seconds.";
   }
-  LOG(ERROR) << "Forward pass: "
-      << static_cast<float>(clock() - forward_start) / CLOCKS_PER_SEC
-      << " seconds.";
-  clock_t backward_start = clock();
+  forward_timer.Stop();
+  LOG(ERROR) << "Forward pass: " << forward_timer.ElapsedSeconds() << " seconds.";
+  Timer backward_timer;
+  backward_timer.Start();
   for (int i = layers.size() - 1; i >= 0; --i) {
     const string& layername = layers[i]->layer_param().name();
-    if (Caffe::mode() == Caffe::GPU) {
-      cudaDeviceSynchronize();
-    }
-    clock_t start = clock();
+    timer.Start();
     for (int j = 0; j < total_iter; ++j) {
       layers[i]->Backward(top_vecs[i], true, &bottom_vecs[i]);
     }
-    if (Caffe::mode() == Caffe::GPU) {
-      cudaDeviceSynchronize();
-    }
+    timer.Stop();
     LOG(ERROR) << layername << "\tbackward: "
-        << static_cast<float>(clock() - start) / CLOCKS_PER_SEC
-        << " seconds.";
+        << timer.ElapsedSeconds() << " seconds.";
   }
-  LOG(ERROR) << "Backward pass: "
-      << static_cast<float>(clock() - backward_start) / CLOCKS_PER_SEC
-      << " seconds.";
-  LOG(ERROR) << "Total Time: "
-      << static_cast<float>(clock() - forward_start) / CLOCKS_PER_SEC
-      << " seconds.";
+  backward_timer.Stop();
+  LOG(ERROR) << "Backward pass: " << backward_timer.ElapsedSeconds() << " seconds.";
+  total_timer.Stop();
+  LOG(ERROR) << "Total Time: " << total_timer.ElapsedSeconds() << " seconds.";
   LOG(ERROR) << "*** Benchmark ends ***";
   return 0;
 }