diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
index 8a92828f5..d09a18f84 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
@@ -192,6 +192,9 @@ static void gk20a_scale_notify(struct platform_device *pdev, bool busy)
 	struct gk20a_scale_profile *profile = g->scale_profile;
 	struct devfreq *devfreq = g->devfreq;
 
+	/* update the software shadow */
+	gk20a_pmu_load_update(g);
+
 	/* inform edp about new constraint */
 	if (platform->prescale)
 		platform->prescale(pdev);
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c b/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
index 994c9cd2f..97a0452cd 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
@@ -278,6 +278,7 @@ static ssize_t gk20a_load_show(struct device *dev,
 		busy_time = 0;
 	} else {
 		gk20a_busy(g->dev);
+		gk20a_pmu_load_update(g);
 		gk20a_pmu_load_norm(g, &busy_time);
 		gk20a_idle(g->dev);
 	}
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
index 7c441f53b..06e7a4e65 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
@@ -3676,17 +3676,23 @@ int gk20a_pmu_destroy(struct gk20a *g)
 }
 
 int gk20a_pmu_load_norm(struct gk20a *g, u32 *load)
+{
+	*load = g->pmu.load_shadow;
+	return 0;
+}
+
+int gk20a_pmu_load_update(struct gk20a *g)
 {
 	struct pmu_gk20a *pmu = &g->pmu;
 	u16 _load = 0;
 
 	if (!pmu->perfmon_ready) {
-		*load = 0;
+		pmu->load_shadow = 0;
 		return 0;
 	}
 
 	pmu_copy_from_dmem(pmu, pmu->sample_buffer, (u8 *)&_load, 2, 0);
-	*load = _load / 10;
+	pmu->load_shadow = _load / 10;
 
 	return 0;
 }
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
index 292aabb0f..694e0288f 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
@@ -1079,6 +1079,7 @@ struct pmu_gk20a {
 	bool perfmon_ready;
 
 	u32 sample_buffer;
+	u32 load_shadow;
 
 	struct mutex isr_mutex;
 	struct mutex isr_enable_lock;
@@ -1119,6 +1120,7 @@ int pmu_mutex_acquire(struct pmu_gk20a *pmu, u32 id, u32 *token);
 int pmu_mutex_release(struct pmu_gk20a *pmu, u32 id, u32 *token);
 int gk20a_pmu_destroy(struct gk20a *g);
 int gk20a_pmu_load_norm(struct gk20a *g, u32 *load);
+int gk20a_pmu_load_update(struct gk20a *g);
 int gk20a_pmu_debugfs_init(struct platform_device *dev);
 void gk20a_pmu_reset_load_counters(struct gk20a *g);
 void gk20a_pmu_get_load_counters(struct gk20a *g, u32 *busy_cycles,