diff --git a/plugins/GSdx/GS.cpp b/plugins/GSdx/GS.cpp
index 28a14f969a..40047b37b8 100644
--- a/plugins/GSdx/GS.cpp
+++ b/plugins/GSdx/GS.cpp
@@ -1528,8 +1528,6 @@ EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
 
 	list<Packet*> packets;
 	vector<uint8> buff;
-	vector<float> stats;
-	stats.clear();
 	uint8 regs[0x2000];
 
 	GSinit();
@@ -1640,11 +1638,8 @@ EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
 		finished = 1;
 	}
 	unsigned long frame_number = 0;
-	unsigned long total_frame_nb = 0;
 	while(finished > 0)
 	{
-		frame_number = 0;
-		unsigned long start = timeGetTime();
 		for(auto i = packets.begin(); i != packets.end(); i++)
 		{
 			Packet* p = *i;
@@ -1686,50 +1681,17 @@ EXPORT_C GSReplay(char* lpszCmdLine, int renderer)
 			}
 		}
 
-		// Ensure the rendering is complete to measure correctly the time.
-		glFinish();
-
 		if (finished >= 200) {
 			; // Nop for Nvidia Profiler
 		} else if (finished > 90) {
 			sleep(1);
 		} else {
-			unsigned long end = timeGetTime();
-			frame_number = std::max(1ul, frame_number); // avoid a potential division by 0
-
-			fprintf(stderr, "The %ld frames of the scene was render on %ldms\n", frame_number, end - start);
-			fprintf(stderr, "A means of %fms by frame\n", (float)(end - start)/(float)frame_number);
-
-			stats.push_back((float)(end - start));
-
 			finished--;
-			total_frame_nb += frame_number;
 		}
 	}
 
-	if (theApp.GetConfigI("linux_replay") > 1) {
-		// Print some nice stats
-		// Skip first frame (shader compilation populate the result)
-		// it divides by 10 the standard deviation...
-		float n = (float)theApp.GetConfigI("linux_replay") - 1.0f;
-		float mean = 0;
-		float sd = 0;
-		for (auto i = stats.begin()+1; i != stats.end(); i++) {
-			mean += *i;
-		}
-		mean = mean/n;
-		for (auto i = stats.begin()+1; i != stats.end(); i++) {
-			sd += pow((*i)-mean, 2);
-		}
-		sd = sqrt(sd/n);
-
-		fprintf(stderr, "\n\nMean: %fms\n", mean);
-		fprintf(stderr, "Standard deviation: %fms\n", sd);
-		fprintf(stderr, "Mean by frame: %fms (%ffps)\n", mean/(float)frame_number, 1000.0f*frame_number/mean);
-		fprintf(stderr, "Standard deviation by frame: %fms\n", sd/(float)frame_number);
-	}
 #ifdef ENABLE_OGL_DEBUG_MEM_BW
-	total_frame_nb *= 1024;
+	unsigned long total_frame_nb = std::max(1ul, frame_number) << 10;
 	fprintf(stderr, "memory bandwith. T: %f KB/f. V: %f KB/f. U: %f KB/f\n",
 			(float)g_real_texture_upload_byte/(float)total_frame_nb,
 			(float)g_vertex_upload_byte/(float)total_frame_nb,
diff --git a/plugins/GSdx/GSDeviceOGL.cpp b/plugins/GSdx/GSDeviceOGL.cpp
index 86f27ead85..70c0050011 100644
--- a/plugins/GSdx/GSDeviceOGL.cpp
+++ b/plugins/GSdx/GSDeviceOGL.cpp
@@ -159,34 +159,41 @@ void GSDeviceOGL::GenerateProfilerData()
 
 	uint64 time_start;
 	uint64 time_end;
+	std::vector<double> times;
+	double ms       = 0.000001;
 
-	uint64 min_time = 0xFFFFFFFFFFFFFFFFull;
-	uint64 max_time = 0;
-	uint64 total    = 0;
+	float replay    = (float)(theApp.GetConfigI("linux_replay"));
+	int first_query = (float)m_profiler.last_query / replay;
 
-	glGetQueryObjectui64v(m_profiler.timer_query[0], GL_QUERY_RESULT, &time_start);
-	for (uint32 q = 1; q < m_profiler.last_query; q++) {
+	glGetQueryObjectui64v(m_profiler.timer_query[first_query], GL_QUERY_RESULT, &time_start);
+	for (uint32 q = first_query + 1; q < m_profiler.last_query; q++) {
 		glGetQueryObjectui64v(m_profiler.timer_query[q], GL_QUERY_RESULT, &time_end);
 		uint64 t = time_end - time_start;
-
-		min_time = std::min(t, min_time);
-		max_time = std::max(t, max_time);
-		total   += t;
+		times.push_back((double)t * ms);
 
 		time_start = time_end;
 	}
 
 	glDeleteQueries(1 << 16, m_profiler.timer_query);
 
-	// FIXME remove 1/linux_replay frame info
-	// Add fps, deviation
-	// remove glFinish
+	double frames    = times.size();
+	double mean      = 0.0;
+	double sd        = 0.0;
 
-	double ms = 0.000001;
-	fprintf(stderr, "\nGenerateProfilerData:\n");
-	fprintf(stderr, "Min  %f\n", (double)min_time * ms);
-	fprintf(stderr, "Mean %f\n", (double)total/(double)m_profiler.last_query * ms);
-	fprintf(stderr, "Max  %f\n", (double)max_time * ms);
+	auto minmax_time = std::minmax_element(times.begin(), times.end());
+
+	for (auto t : times) mean += t;
+	mean = mean / frames;
+
+	for (auto t : times) sd += pow(t-mean, 2);
+	sd = sqrt(sd / frames);
+
+	fprintf(stderr, "\nPerformance Profile for %.0f frames:\n", frames);
+	fprintf(stderr, "Min  %4.2f ms\t(%4.2f fps)\n", *minmax_time.first, 1000.0 / *minmax_time.first);
+	fprintf(stderr, "Mean %4.2f ms\t(%4.2f fps)\n", mean, 1000.0 / mean);
+	fprintf(stderr, "Max  %4.2f ms\t(%4.2f fps)\n", *minmax_time.second, 1000.0 / *minmax_time.second);
+	fprintf(stderr, "SD   %4.2f ms\t(%4.2f fps)\n", sd, 1000.0 / sd);
+	fprintf(stderr, "\n");
 }
 
 GSTexture* GSDeviceOGL::CreateSurface(int type, int w, int h, bool msaa, int fmt)